Science 1860 2008

Volume 321, Issue 5894

COVER

Mammalian fatty add synthase, a
multienzyme that catalyzes all steps of fatty
acid biosynthesis. A blueprint of its atomic
structure is shown in three views, and the
extent of its functional domains is indicated
by colored bars. The versatile segmental
construction is also used in other members
of this large family of multienzymes,
which synthesize natural products such
as antibiotics. See page 1315.

Image : Marc Leibundgut and Timm
Maier/ETH Zurich

DEPARTMENTS

1267 Science Online

1 268 This Week in Science
1272 Editors' Choice

1274 Contact Science

1275 Random Samples
1277 Newsmakers

1 366 Gordon Research Conferences

1374 New Products

1375 Science Careers

EDITORIAL

1271 Scientific Publishing Standards
by Bruce Alberts

NEWS OF THE WEEK

Whole-Genome Data Not Anonymous, Challenging 1278
Assumptions

China Plans $3.5 Billion GM Crops Initiative
A Detailed Genetic Portrait of the Deadliest
Human Cancers

» Science Express Research Articles by 0. W. Parsons et al.
and S. fones et at

Hippocampal Firing Patterns Linked to Memory Recall 1280

» Science Express Report by H. Gelbard-Sagtv et aL;

Research Article p. 1322

SCIENCESCOPE 1281

MathFest 2008 Meeting
Shapeshifting Made Easy
Sweet inspiration
A Royal Squeeze
Taking the Edge Off

NEWS FOCUS

Investigating the Psychopathic Mind

» Science Podcast

Large Hadron CoWder

The Overture Begins 1287

Researchers, Plate Your Bets!

Bracing for a Maelstrom of Data, CERN Puts Its Faith 1 289

in the Grid

Is the LHC a Doomsday Machine?

LETTERS

Reading Between the Number Lines R. E. Nunez 1293
Response I/. Izard, S. Dehaene, P. Pica, E. Spelke
The Risks of Pigging Out on Antibiotics

R. Goldburg, S. Roach, D. Wallinga, M. Mellon
Battle of the Bugs R. D. Sleator and C. Hill
DOE Should Keep Education in Mind L. A Kult
Call for an Objective DOE Decision C Cassapakis
CORRECTIONS AND CLARIFICATIONS 1295

BOOKS ETAL

Doubt Is Their Product How Industry’s Assault
on Science Threatens Your Health
D. Michaels, reviewed by C. F. Cranor
A Taste of the Gonzo Scientist

» Online feature p. 1267

POLICY FORUM

Life Cycle of Translational Research for
Medical Interventions

D. G. Contopoulos-loannidis et al.

PERSPECTIVES

Enhancing Gene Regulation

G. A YJray and C C. Babbitt
» Brevia p. 1314; Report p. 1346

The Universe Measured with a Comb

S. Lopez •> Report p. 1335

The Cart Before the Horse
]. D. Rowley and T. Blumenthal
An Enzyme Assembly Line
y. L Smith and D. H. Sherman
» Research Article p. 1315
How to Infect a Mimivirus

H. Ogata and J.-M. Ctaverie
An End to the Drought of Quantum Spin Liquids
PA. Lee

CONTENTS continued »

1296

1297

1300

1301

1302

1304

1305

1306

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

CONTENTS

SCIENCE EXPRESS

www.sciencexpress.org

NEUROSCIENCE

Internally Generated Reactivation of Single Neurons in

Human Hippocampus During Free Recall

H. Gelbard-Sagiv, R. Mukamel, M. Hard, R. Malach, I. Fried

The firing patterns ol brain neurons recorded from people watching a video episode

were the same as those recorded during later recall of the same show.

» News story p. 1280; Research Article p. 1322

10.1126/science.ll64685

CHEMISTRY

Merging Photoredox Catalysis with Organocatalysis: The Direct Asymmetric

Alkylation of Aldehydes

D. A Nicewia and D. W. C. MacMillan

When irradiated by light, a ruthenium-organic catalyst creates intermediates with
unpaired electrons that undergo otherwise intractable asymmetric reactions.

10.1 12 6/science. 1161976

CELL BIOLOGY

TMEM16A, A Membrane Protein Associated with Calcium-Dependent
Chloride Channel Activity
A Caputo et aL

A transmembrane protein induced in cytokine-treated bronchial epithelial cells seems to
be a long-sought primary carrier of a voltage- and calcium-dependent chloride current.

10.1126/science.ll63518

MEDICINE

An Integrated Genomic Analysis of Human Glioblastoma Multiforme
D. I V. Parsons et al.

Comprehensive analysis ol mutations in a brain cancer identifies previously unrecognized
cancer genes and a frequently mutated protein that may serve as a therapeutic marker.

» News story p. 1280; Science Express Research Article by 5. Jones et al.

10. 1126/science. 1164382

MEDICINE

Core Signaling Pathways in Human Pancreatic Cancers Revealed by
Globa l Genomic Analyses

S. Jones et al.

Analysis of genome alterations shows that the same 12 signaling pathways are
disrupted in most pancreatic tumors, suggesting these as key to tumor development.

» News story p. 1280; Science Express Research Article by D. I/K Parsons et at.

10. 1126/science. 1164368

TECHNICAL COMMENT ABSTRACTS

ECOLOGY

Comment on "Fire-Derived Charcoal Causes Loss 1295
of Forest Humus"

]. Lehmann andS. Sohi

full text at mrw.sciencemag.org/cgi/contenVfaH/321/S894/129Sc

Response to Comment on "Fire-Derived Charcoal
Causes Loss of Forest Humus"

D. A. War die, M.-C. Nilsson, 0. Zackrisson

full text at www.saencemag.org/cgi/contenVfull/321/S894/1295d

REVIEW

ATMOSPHERIC SCIENCE

Flood or Drought: How Do Aerosols
Affect Precipitation?

D. Rosenfeld et al.

BREVIA

DEVELOPMENTAL BIOLOGY

Shadow Enhancers as a Source of 1314

Evolutionary Novelty

J.-W. Hong, D. A. Hendrix, M. S. Levine

Some developmental important genes can be regulated via two

enhancers, one located nearby and the other, a “shadow" enhancer,

10 to 20 kilobases away.

» Perspective p. 1300; Report p. 1346

RESEARCH ARTICLES

STRUCTURAL BIOLOGY

The Crystal Structure of a Mammalian 1315

Fatty Acid Synthase

T. Maier, M. Leibundgut, N. Ban

A high-resolution structure of mammalian fatty acid synthase reveals
that this enzyme is derived from an iterative polyketide synthase and
has five active catalytic domains. » Perspective p. 1304

NEUROSCIENCE

Internally Generated Cell Assembly Sequences 1322
in the Rat Hippocampus

E. Pastalkova, V. Itskov, A. Amarasingham, G. Buzsaki
As rats perform a memory task, cells in their hippocampus fine
in sell-generated sequences that correspond to and presage the
animals' subsequent choices. » News story p. 1280; Scii>n.

Report by H. Gelbard-Sagiv et al»* Science Podcast

REPORTS

GEOCHEMISTRY

Experimental Test of Self-Shielding in Vacuum 1328
Ultraviolet Photodissociation of CO
S. Chakraborty, M. Ahmed, T. L. Jackson, M. H. Thiemens
The anomalous variation of oxygen isotopes in early meteorites
is produced by excited states during photodissociation of carbon
monoxide, not by self-shielding, as was thought.

CONTENTS continued »

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1263

Science

CONTENTS

REPORTS CONTINUED...

CHEMISTRY

Identification of Active Gold Nanoclusters on
Iron Oxide Supports for CO Oxidation
A. A Herring et aL

High-resolution microscopy showed that the most effective catalytic
gold species on an iron oxide support were those forming bilayer
clusters of just 10 atoms.

ASTRONOMY

Laser Frequency Combs for Astronomical

Observations

T. Steinmetz et al.

Accurate spectroscopy of the sun with a laser frequency comb shows
that it can improve astronomical observations and may yield direct
evidence of the universe's expansion.

» Perspective p. 1301

PALEOCLIMATE

Regional Synthesis of Mediterranean Atmospheric 1 338
Circulation Outing the Last Glacial Maximum
). Kuhlemann et al.

A three-dimensional reconstruction of atmospheric temperatures in
the Mediterranean during glacial times is analogous to one of winter
during the Little Ice Age.

CLIMATE CHANGE

Kinematic Constraints on Glacier Contributions to 1 340
21st-Century Sea-Level Rise
W. I Pfeffer, J. T. Harper, S. O'Neel

Evaluation ol glacier dynamics implies that melting of the Greenland
and Antarctic Ice Sheets could raise sea level by up to 2 meters by
2100, although a rise of 0.8 meters is more likely.

» Science Podcast

IMMUNOLOGY

Apobec3 Encodes Rfv3, a Gene Influencing 1343

Neutralizing Antibody Control of Retrovirus Infection
AL L. Santiago et al.

A resistance factor known to protect mice from retroviral infection is
unexpectedly identified as Apobec3, a deoxycytidine deaminase that
controls somatic hypermutation.

GENETICS

Human-Specific Gain of Function in a
Developmental Enhancer
S. Prabhakar et al.

When transferred to a mouse, a conserved regulatory element that
has been positively selected in humans is robustly expressed at the
base of its developing thumb and wrist.

» Perspective p. 1300; Brevia p. 1314

CELL BIOLOGY

Wnt3a-Mediated Formation of Phosphatidylinositol 1350
4,5-Bisphosphate Regulates LRP 6 Phosphorylation
W. Pan et al.

The interaction of the signaling molecule Wnt to its receptor triggers
accumulation of a lipid regulator, which stimulates phosphorylation
of the receptor and cellular responses.

BIOCHEMISTRY

Helical Structures of ESCRT-III Are Disassembled 1354
by VPS4
5. Lata et al.

A protein responsible for the final separation of daughter cells
or budding viruses forms heteromeric complexes on the inside
of the membrane to regulate the abscission step.

MEDICINE

A Neoplastic Gene Fusion Mimics Trans-Splicing of 1357
RNAs in Normal Human Cells
H. Li l Wang, G. Mor, ). Sklar
A chimeric messenger RNA generated in a tumor by a DMA
rearrangement is also, unexpectedly, expressed in healthy cells,
a result of splicing together two separate messenger RNAs.

» Perspective p. 1302
MEDICINE

Germline Allele-Specific Expression of TGFBR1 1 361
Confers an Increased Risk of Colorectal Cancer
L Valle et al.

In patients with colorectal cancer, one allele of the transforming
growth factor— p gene produces less messenger RNA and thus less
protein, a likely contributor to disease risk.

SaCHCt (ISSN 00HHP7SI ll pobtbkrd noddy M fridoy. oittp * tbe Ult nook in Dncombor, by tho Amt firm AttorlMioo
(•! rh* Advoivcrmtirt ol srloorr, 1 ?oo Now York Avow t. NW, Mwkiogtoo, DC 7000S. Porteiioli Mot pottogo IpublkoCkmNa.
4P446M poid it VbiMngtwi. DC. ond oddttonoi molbig oflfcw. Copyright O 200* by tho Amt* on Auocbtlon for the Adromnmo*
t# Sdtrat. Thottto SCCNCE b o'tfpitmdtndtnu* id Bit AMS. Cairo, tk IrtiMdiul monboriMpond lubtrrkxkn rSlbom): s J 44
(S74 oUocitod to ubicrtpdon). DowHtk butturtonol tabtcrlptlon (SI ImimJ: S 770; rordpn postoao «tn: Mtrko, CodbOton Csurfoco
ADVANCING SCIENCE. SERVING SOCIETY m«C SSS; alhor towitrlostokottld dtdvtry) S«S. Fbst dots, olrooll.itildwit. ond omrrttiu ratti on roqdost Conation rotomtti 051
ovalibto «pon rt»itst, « ! M2S4 88122. PybbcotSom Mod AprttmtM Nwnbor 1060424. SCIENCE k prWtd on TO perant past
twuimtr rtcydtd popon Priatrd i n tbt US A.

Choog'ofoMrcic Allow Iwodu.^thg old ond muMuis ond tdtptoaoM rnmbw. Pooanorttr Sood .honpoof oddrotstoAAAS, f.O. 8oit041 7*. Wmhbgton. DC 2C3W-4i;8 S(oglo*opyiolci:
SIOCIO award I wo. S1S.00 bode tuur ptapoM trtdpWi nrfoco pottage bidtratis on rttpj-* AotborlatSoo to pbotonpy mitorltl for Irttmol or ptrunol lilt ondtr dtenuamn not (Ollbg idcMn tto
fob ico provUm of tht Copy rfgtt A a Is grinttd by AAA S to U>ro*n ord oditr turn rogktwid nth tho Copjtkjhr Clnironco Corttr (CCO Traruocdoml loporSng SuvStr. prorUod But S20.00 ptr odlcit b
paid dkodty to CCC. 222 Aotmood Orlro, Donton, MA 0192 J. tho tdtfdffcodon cadrforSd*wrb(XIJ6-M7S.S<boc4likdoj»dk,th>fdodw'(<bd<Srf,>AmadKaft*«vftororidl,iwtral tporblbodbdraot.

o

Printed on

30% post-consumer
recycled paper.

CONTENTS continued »

www.sciencema 9 . 0 rg SCIENCE VOL 321 5 SEPTEMBER 2008

1265

www.sciencemag.org

{

Skeletal development requires the CaSR.

SCIENCE SIGNALING

www.sciencesignaling.org

THE SIGNAL TRANSDUCTION KNOWLEDGE ENVIRONMENT

EDITORIAL GUIDE: Seeing the Signaling Forest and the Trees

M. B. Yaffe

Science Signoiing launches primary research to meet the needs of the signal
transduction community.

Development

RESEARCH ARTICLE: The Extracellular Calcium-Sensing Receptor (CaSR)

Is a Critical Modulator of Skeletal Development

W. Chang, C Tu, T.-H. Chen, D. Bikie, D. Shoback

PERSPECTIVE: New Insights in Bone Biology— Unmasking Skeletal Effects

of the Extracellular Calcium-Sensing Receptor

E. M. Brown and). B. Lian

The extracellular calcium-sensing receptor (CaSR) is essential for embryonic and
postnatal skeletal development.

RESEARCH ARTICLE: Linear Motif Atlas for Phosphorylation-Dependent Signaling

M. L Miller, L ). Jensen, F. Dielta, C. jargensen, M. Tinti, L Li, M. Hsiung,

S. A. Parker, ). Bordeaux, 1 Sicheritz-Ponten , M. Olhovsky, A Pascutescu,

). Alexander, S. Knapp, N. Blom, P. Bork, S. Li G. Cesareni, 1 Pawson,

B. E. Turk, M. B. Yaffe, S. Brunak, R. Linding

Created with both in vitro and in vivo data, NetPhorest is an atlas of consensus
sequence motifs for 179 kinases and 104 phosphorylation-dependent binding
domains and reveals new insight into phosphorylation-dependent signaling.

REVIEW: Alternative Wnt Signaling Is Initiated by Distinct Receptors
R. van Amerongen, A Mikels, R. Nusse

Tire traditional classification of Writs into canonical or noncanonical proteins may
^ be misleading.

§ SCIENCE ONLINE FEATURE

THE GONZO SCIENTIST: How Astronomers
Have Fun (and Nearly Die Trying)
g In western Mongolia, a sola r eclipse has

mythic meaning (with audio slideshow).

- wvm.sciencemag.org/sdext/gon/oscientiit/

SCIENCENOW

www.sciencenow.org

HIGHLIGHT S FROM OUR DAILY NEWS COVERAGE

Taking One for the Team

Selflessness might be bad for the warrior but good

for the tribe.

Fancy Footwork Helps Flies Cheat Death
High-speed videos reveal surprising sophistication
in insect's escape response.

Why Men Cheat

Study chalks up promiscuous behavior to a single genetic
change.

A particle physicist at the
Large Hadron Collider.

SCIENCE CAREERS

www.sciencecareers.org/career_development

FREE CAREER RESOURCES FOR SCIENTISTS

Working in Industry: Taken for Granted —

Fitting the Job Market to a T

B. Benderly

Scientists need more than bench expertise to find work in industry.
Working in Industry: Mastering Your Ph.D. — Is Industry
Right for You?

B. Noordam

Research in industry differs from academic research in several ways.
Triggermeister

C. Reed

Particle physicist Bitge Demirkoz will make sure colleagues see what
happens when CERN's Large Hadron Collider starts this month.

September 2008 Funding News

]. Fernandez

Learn about the latest in research f unding, scholarships, fellowships,
and internships.

SCIENCE PODCAST

www.sciencemag.orgraboutrpodcast.dtt

FREE WEEKLY SHOW

Download the 5 September
Science Podcast to hear
about organizing memory
in the hippocampus, ice loss
and sea level rise, criminal
psychopathy, and more.

e

Separate individual or institutional subscriptions to these products may be required for fuU-text access.

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1267

EDITED BY STELLA HU RTLEY

Toward Precision Astronomy »

In a dynamical universe (one that is expanding), you would
expect to see Doppler, or red shifts, of spectrograph lines of
distant objects that are moving further and faster away from
the observer, and for such shifts to drift in velocity over time.
Determining the extent of velocity drift, however, requires a
level of precision that has not been available — less than one
centimeter per second per year. Steinmetz et at. (p. 1335; see
the Perspective by Lopez) show how that situation may change
using a laser frequency comb produced by an optic fiber. In a
proof-of-principle experiment, they combine such a comb of
equally spaced wavelengths with an astronomical observation
(the Sun) to precisely measure and calibrate the wavelengths
of the spectrogram.

Moderating Rainfall

Aerosols can either increase or decrease rainfall,
so why do they act sometimes one way and
sometimes another? Rosenfeld et at. (p. 1309)
review the role of aerosols as moderators of pre-
cipitation, and propose a conceptual model to
explain their apparently contradictory effects.
Even small amounts of aerosols in very clean air
prevent the development of long-lived clouds
that can deliver large amounts of rain, whereas
heavily polluted clouds evaporate much of their
water before they can rain through a combina-
tion of microphysical and radiative effects. Thus,
precipitation occurs most efficiently and abun-
dantly at moderate aerosol concentrations.

Back in Circulation

The climate of the Mediterranean region during
the Last Glacial Maximum, between 23,000 and
19,000 years ago, is known to be much colder
than today, but the atmospheric circulation pat-
terns that prevailed remain poorly understood.
Kuhlemann et aL (p. 1338, published
online 31 July) synthesized a range of
new and published data on the equi-
librium line altitude of glaciers (the
altitude at which ice covers the
ground all year long), paleoflora, and
regional sea surface temperatures,
and reconstructed the three-dimen-
sional temperature structure of the
atmosphere. Atmospheric circulation was
like that observed commonly in the winters of the
Little Ice Age, roughly between 1500 and 1900.

Putting Limits on Ice Loss

Ice loss from the margins of the Greenland and
Antarctic ice sheets can occur through dynami-

cally forced discharge from fast flowing ice
streams and calving of marine-terminating gla-
ciers. However, so little is known about ice sheet
dynamics that models are unable to represent
these processes accurately. Instead of trying to
add up estimates of individual source contribu-
tions, Pfeffer et at. (p. 1340) calculated how
much ice discharge from outlet glaciers in
Greenland and Antarctica would be required to
produce various rates of sea level rise, and then
evaluate the plausibility of those discharge
rates. Estimates of more than 2 meters of sea
level rise by 2100 are highly unlikely — a more
reasonable estimate is between 80 centimeters
and 2 meters.

Coding Space, Time,
and Memory

Mental operations such as planning, free recall,
and problem-solving are assumed to depend on
the central nervous system's self-organized

sequences of activity,
which permit cognitive
representations, in
sequence, of the future
or the past. Similar cog-
nitive content should be
represented by similar
assembly sequences,
and different content
should be distinguished by
distinctive sequences. Experimental verification
of this hypothesis has had to wait for large-scale
assembly recordings. Pastalkova et aL (p. 1322;
see the News story by Miller) report that, during
the delay period of a memory task when an ani-
mal is running in a running wheel, each time
point is characterized by the firing of a particular

constellation of hippocampal neurons that
form a highly specific activity sequence across
time. During learning, the temporal order of
multiple external events is instrumental in select-
ing the appropriate neuronal representations,
whereas, during free recall or action planning,
the intrinsic dynamics of the hippocampal system
determines sequence identity.

Focus on Fatty Acid
Synthase

Structural studies have led to an increased under-
standing of the large enzyme systems responsible
for the synthesis of fatty acids, polyketides, and
nonribosomal peptides. Now Maier etaL (p. 1315,
see the cover and the Perspective by Smith and
Sherman) report a structure of porcine fatty acid
synthase (FAS) that includes five of seven cat-
alytic domains, two nonenzymatic domains, and
various linkers. The structure shows how the
linker regions and catalytic domains are organ-
ized to provide the flexibility required for itera-
tive fatty acid elongation. Like modular polyke-
tide synthase, mammalian FAS acts as a
"megasynthase" that can accommodate insertion <
or deletion of product modifying domains to |

allow generation of diverse products. |

t

Tracking Evolution of |

Transcription Regulation i

Bioinformatic approaches are providing insight 1
into the evolution of noncoding regulatory ele- 2
ments of genes that can drive different expres- £
sion outcomes from similar sets of genes (see e
the Perspective by Wray and Babbitt). Several i
recent computational efforts have identified 8

1268

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

This Week in Science

conserved noncoding sequences that have evolved rapidly in humans, but it is not known whether
their functions might have changed during the evolution process. Prabhakar et aL (p. 1346) used
such a noncoding element called human-accelerated conserved noncoding sequence 1 (HACNS1),
as well as orthologs of the gene from nonhuman primates, to create transgenic mouse embryos.
HACSN1, but not the nonprimate orthologs, drove expression of a reporter gene at the junction of
the anterior developing hand and forearm, including the base of the developing thumb and wrist.
Sequence changes were identified that could "humanize" the expression patterns of the chim-
panzee enhancer. Hong et al. (p. 1314) searched for clusters of potential transcription factor
binding sites, in this case for targets of regulation by the transcription factor Dorsal and known
cofactors in the fruit fly. Some of the secondary, or "shadow," enhancers have patterns of gene
expression that overlap those of primary enhancers, and may be able to evolve without disrupting
core expression patterns.

ESCRTing Membrane Scission

So-called ESCRT proteins have been implicated in catalyzing different cellular and
pathological processes, including multivesicular body biogenesis, retrovirus bud-
ding, and cytokinesis. These processes involve topologically similar membrane
events that require a common final abscission step to separate two newly formed
membrane-enveloped structures. Little is known regarding how the budding
steps, including membrane abscission, are catalyzed. Indirect evidence suggests
that ESCRT-III plays an important role in the final step. Notably, dominant-negative
CHMP3, a subunit of ESCRT-III, inhibits HIV-1 budding as well as cytokinesis.

Because cytokinesis does not require vesicle formation, it would seem that ESCRT-III
regulates steps in membrane abscission. Lata etai (p. 1354, published online
7 August) provide structural evidence for the formation of distinct heteromeric
ESCRT-III assemblies by electron microscopy. These structures could bind on the
inside of the neck of a bud or at the midbody between dividing cells and regulate
membrane abscission.

The Normal Side of Trans-Splicing

Human tumors frequently display chromosomal rearrangements that fuse two distinct genes and result
in the expression of chimeric messenger RNA (mRNA) transcripts whose protein products are oncogenic.
Li etal. (p. 1357; see Perspective by Rowley and Biumenthal) suggest that the chimeric mRNAs gen-
erated by chromosomal rearrangements in tumors may sometimes represent constitutively expressed
versions of chimeric mRNAs generated in healthy tissue by trans-splicing. Studying a )AZF1-)JAZ1
chimeric transcript that is abundantly expressed in human endometrial stromal sarcomas with a (7;17)
chromosomal translocation, the authors found unexpectedly that the same chimeric transcript was
expressed in normal endometrial stromal cells, even though these cells tacked the chromosomal trans-
location. In normal cells, the chimeric transcript arose by trans-splicing between independently tran-
scribed JAZF1 and ]JAZ1 pre-mRNAs, and it was translated into a chimeric protein of unknown function.
Trans-splicing is thought to be a rare event in mammalian cells, but these results suggest that other
examples might be found by searching for normal RNA counterparts to the many chimeric mRNAs gen-
erated by chromosomal rearrangements in tumors.

Quantity, Not Just Quality, Matters

Colorectal cancer is one of the leading causes of cancer-related deaths worldwide. Because 20 to 30%
of cases occur in individuals with a family history of the disease, genetic factors are thought to be a
substantial contributor to risk. Valle et al. (p. 1361, published online 14 August) now report that one
of these factors is an inherited variation in the expression level of a gene encoding a key signaling pro-
tein previously implicated in colorectal cancer pathogenesis. Within a Caucasian population in the
United States, individuals with colorectal cancer are 5 to 10 times more likely than controls to show
germline allele-specific expression of the TGFBR1 gene, which encodes the type I receptor for trans-
j forming growth factor-(J. Allele-specific expression appears to result in a modest, but biologically
fc meaningful, lifelong reduction in the expression of TGFBR1, which in turn confers an increased risk of
§ colorectal cancer. Thus, it seems that the genetic contribution to disease risk includes not only muta-
| tions that abolish or modify the function of genes but also more subtle alterations that change the
<5 baseline expression levels of genes.

Try the new
lnvitrogen.com

i invitrogen*

www.invitrogen.com

OXXM InmuoynCorportton M rqtarts

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

Scientific Publishing Standards

THIS WEEK MARKS THE LAUNCH OF A NEW, IMPROVED VERSION OF THE AAAS JOURNAL
Science Signaling, which will now contain original research. Professor Michael Yaffe, its new
Bruce Alberts is the chief scientific editor, has clearly articulated the ambitious goals for his journal.* Highly elabo-

Editor-in-Chief ot Science, rate signaling mechanisms are essential for controlling the behavior of each cell in a multicellu-

lar organism — allowing each of the many billions of cells in our bodies to decide whether it will
grow and proliferate, remain quiescent, kill itself, or change its behavior according to signals
received from neighboring cells. Understanding how this complex system works represents a
major challenge. Unraveling its many mysteries will require a great deal of ingenuity — and the
collaboration of biologists, chemists, physicists, engineers, computer scientists, and mathemati-
cians. We are confident that Science Signaling will set the highest standard for research in this
important ft eld, and that, through its Perspectives and Review articles, it will help to guide future
researchers along highly productive paths.

The new journal began in 1999 as Science's STKE (Signal Transduc-
tion Knowledge Environment), an online resource. The initial aim was to
speed the generation of new knowledge by creating an Internet-based
work environment that would provide “all practitioners in a field of
endeavor access to all the knowledge within the field” and “speed iden-
tification of relevant information and encourage communication with
others.” The Web site flourished. As the next step in an ongoing evolu-
tion, Science Signaling has now added original peer-reviewed research
papers to the myriad of resources provided at the site.

I want to take this propitious occasion to reflect briefly on the core
purpose of scientific publishing, and to consider some guiding principles
that we scientists, editors, and publishers need to keep in mind in our col-
lective efforts to improve the scientific literature.

The publication of a scientific article is less a way for scientists to earn recognition and
advance their careers than it is an engine for scientif c progress. Science continually advances
only because many cycles of independent testing by different scientists allow new knowledge to
be built with confidence upon old knowledge, thereby creating a repository of reliable under-
standings about the world. The publications of those of us who are scientists explain what we
have found in our investigations, and they lay out exactly what we have done to make each dis-
covery. Clear, truthful presentations of data, results, and methods are essential for enabling the
findings of one scientist to be confirmed, refuted, or extended in new ways by other scientists.

Scientists have an absolute obligation to honesty: They must accurately report how they
arrived at their discoveries, as well as the discoveries themselves. Thus, our journals must insist
on detailed descriptions of all of the methods used, so as to allow other scientists to reproduce
the results in a straightforward manner. The appropriate place for most of this information is in
the easily expandable Supplementary Materials that accompany each article. Authors, review-
ers, and editors of scientific manuscripts should therefore constantly ask themselves whether the
reader has been provided with everything needed to both understand and reproduce the results.

The increasingly large data sets produced in some studies present a different challenge; they
require deposition in readily accessible, online archives, supported by stable public funding.

Last but not least, journals themselves can certainly set a higher bar for the clarity of presen-
tation in the manuscripts that we publish. The problem is perhaps most obvious in the brief
abstracts that authors write to introduce each article, which often seem to be written only for a
handful of experts in the authors’ subspecialty. Some abstracts, full of three-letter abbreviations
and jargon, are incomprehensible to me even in my own field of cell biology. As scientists and
as journal publishers, we can and we must do better. In this, as in many other areas, Science
Signaling will aim for the same high standards that we strive for at Science.

-Bruce Alberts

10.1126/sc ence.!16S268

•Sri Signal. 1, eg5 (2008).

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1271

EDITED BY GILBERT CHIN AND JAKE YESTON

PHYSICS

Snapshot Magnetometry

In cold-atom chips, atoms are guided above tracks
of wires that supply the magnetic field to keep
them aloft. In applications ranging from quantum
information processing to metrology, any devia-
tion in the magnetic field from point to point over
the chip could influence the delicate state of the
atoms. Terraciano etaL introduce a technique
that takes a snapshot image of the magnetic field
landscape. Using a cloud of cold rubidium atoms,
whose energy levels are sensitive to magnetic
field, they let the cloud fall toward the chip and
probe the atoms' state with a laser beam tuned to
one of the magnetic transitions. The ability to
take a two-dimensional snapshot image of mag-
netic field variations of 30 mG/cm above the
atom chip over 5 mm with 250-pm resolution
should prove useful in calibrating these chips for
their envisioned applications. — ISO

Opt. Express 16, 13062 (2008).

BIOCHEMISTRY

Translation Translocations

Ribosomes translate mRNA into protein with the
help of GTPases: the elongation factors (EFs). In
prokaryotes, as each mRNA codon is presented in
the A site of the ribosome, EF-Tu loads a comple-
mentary, amino acid-bearing tRNA into the A site.
After peptide bond formation, EF-G translocates
the ribosome along the mRNA strand by three
nucleotides, moving the tRNA (now carrying the
nascent polypeptide chain) into the neighboring P
site and bringing the next codon into the A site.
The GTPase EF4/LepA was recently found to pro-
mote backward translocation of the ribosome
along the mRNA strand, moving the tRNA from

the P site back into the A site. This func-
tion may allow the ribosome to
recover from forward transloca-
tions of the wrong number
of nucleotides. Connell et
a(. have visualized EF4
in complex with a ribo-
some and associated
tRNAs using single-parti-
cle cryo-electron microscopy
vOC (EM). Fitting the crystal struc-
ture of EF4 into the cryo-EM
EF4 (red) grabs reconstruction revealed that
the A-site tRNA its C-terminal domain forms
(purple). multiple contacts with a tRNA

•Nilah Mornier is a summer intern in Science's editorial
department.

GEOPHYSICS

Sensing Supershear

Recent observations, supported by experiments, have indicated that some earthquake ruptures
transiently exceed the local speed of sound along the fault zone. This "supershear" can explain
enhanced shaking from these quakes; thus, supershear ruptures are critical in assessing seismic
risks. Many of the details of how ruptures accelerate to above the sound speed and then decelerate,
in some cases repeatedly, as a rupture progresses are unclear, as most supershear ruptures have
been inferred by data inversions. Vallee et al. were able to observe these dynamics more directly in
the 2001 Kokoxili earthquake = 7.8) — which ruptured 400 km along the Kunlun fault in north-
ern Tibet — thanks to an array of seismometers in Nepal that were nearly parallel to the rupture.
Their data show that the earthquake, which began in the west, accelerated to above the shear wave
velocity after ripping 175 km eastward, at a bend in the fault Rupture speeds nearly reached the
compressional (p) wave velocity before decelerating at another bend. Much of the high-frequency
seismic energy from the quake was radiated during these transitions. — BH

]. Geophys. Res. 113, B07305 (2008).

in the A site, suggesting that EF4 promotes back-
translocation by stabilizing the A-site tRNA posi-
tion over the P-site tRNA position. — NM*

Nat. Struct. Mol. Biol. 15, 10.1038/
nsmb.1469 (2008).

CLIMATE SCIENCE

1000 Years of Hurricanes

The natural variability of hurricane activity is
poorly known, not least because the historic
record for hurricanes extends back only about
130 years. As a result, there has been controversy
over whether hurricane activity will change — or

is already changing — as a result of global warm-
ing. Sediments may hold clues to hurricane activ-
ity over longer time scales, but few studies have
yielded sedimentary records of hurricane activity
at annual resolution. Besonen etal. have now
obtained an annually resolved lake sediment
record from Lower Mystic Lake in Boston, Massa-
chusetts, that covers the past 1000 years. The
record contains anomalous features — unusually
thick layers in which coarse sediments and terres-
trial, organic detritus are overlain by progres-
sively finer sediments — that are indicative of
strong flooding. Comparison with the historic
record shows that 10 out of 11 of these features

2

8

i

I

8

1272

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

EDITORS'CHOICE

i

S

si

s

1

I

i

3

occur in years when category 2-to-3 hurricanes
struck Boston. The authors use this correlation to
determine centennial-scale changes in hurricane
frequency. Further records of this type from other
locations will help to relate these patterns to
other paleoclimate indicators. — JFU

Geophys. Res. Lett. 3S, 114705 (2008).

BIOCHEMISTRY

Plasmid Propulsion

To be propagated stably in prokaryotes, low-copy
number plasmids must be allocated actively dur-
ing cell division. The R1 plasmid is maintained at
four to six copies per cell by the par operon, which
encodes the DNA-binding protein ParR and the
actin-like ATPase ParM. ParR binds cooperatively
as a dimer to 11-base pair repeats in
parC ; ParM undergoes ATP-depend-
ent polymerization, but only grows
into long parallel filaments that are
capable of pushing replicated plas-
mids apart when capped by the
ParR-po/C complex. To understand
how elongating filaments are stabi-
lized, Saljeand

Lowe have used ParM (yellow)

electron microscopy capped by ParR

and biochemistry to (purple),

determine the archi-
tecture of capped filaments. ParR -port
complexes have previously been shown to
form a clamplike structure in which parC
DNA wraps around a helical array of ParR
dimers. Guided by biochemical mapping of the
ParR-ParM interaction sites, they modeled the
crystal structure of ParR onto the end of the dou-
ble-helical ParM filament. The ParR -parC clamp
wraps around the filament with the C-terminal
regions of ParR bound to exposed loops of ParM.
Each ParR -parC complex binds the end of a single
filament, and the filament ends can be bound
simultaneously. Unlike actin, ParM forms left-
handed filaments, which allows ParM monomers
access to the ends of protofilaments capped with
right-handed ParR -parC. The authors suggest a
model in which force is produced by the alternat-
ing addition of monomers to each protofilament
accompanied by rocking of the ParR clamp from
side to side, analogous to the model proposed for
formin-assisted actin polymerization. — VV

EMBO ]. 27, 2230 (2008).

BIOPHYSICS

Molecular Cloaking

Natural products, such as latex rubber or beta-
lactam antibiotics, have given rise to entire
industries, and green fluorescent protein (GFP)
has fought its way onto the list. A series of vari-

ants created in several laboratories have shifted
the peak excitation and emission wavelengths
(for multicolor imaging), improved the photosta-
bility (for time-lapse cinematography), and
enhanced the quantum yields (lowering detec-
tion thresholds). Andresen et al. describe their
latest entry — which has been christened Padron
in recognition of its "reversed" behavior in com-
parison to its parent, Dronpa — and demonstrate
howto implement multilabel, single-color imag-
ing. Dronpa and its widely used descendant
rsFastLime fluoresce when excited with blue light
(488 nm), which also converts them from an
"on" state to an "off" or nonfluorescent state,
from which they can be switched on again by
irradiation with ultraviolet (UV) light (405 nm).

In contrast, Padron (differing at eight amino acid
residues from Dronpa) is switched off by UV
and on by blue light. As the emission of
both proteins is centered at roughly
520 nm, and both exhibit very low off-
state fluorescence, a single detection
window can be used. — GJC

Nat. Biotechnol. 26, 10.1038/
nbt.1493 (2008).

BIOMEDICINE

From Clinic to Lab and Back

Some breast cancer patients respond to doc-
etaxel chemotherapy, but some do not.

Honma etal. have marshaled converging evi-
dence that ribophorin II (RPN2), a mammalian
oligosaccharide transferase component con-
tributes to the development of resistance to doc-
etaxel. Assessing gene expression levels in non-
responders versus responders yielded 85 genes
expressed at higher levels in nonresponsive
patients. Down-regulating these genes individu-
ally by applying small interfering RNAs (siRNAs)
to a docetaxel-resistant breast cancer cell line win-
nowed the candidates to eight, with RPN2 knock-
down strongly associated with the inhibition of
cell growth (taxanes are antimitotic agents) and
activation of apoptotic (programmed cell death)
pathways; conversely, docetaxel-resistant cells dis-
played enhanced expression of RPN2 and also of
MDR1, which encodes a multidrug efflux pump.
Translating these findings into two animal mod-
els — created by implanting two docetaxel-resist-
ant breast cancer cell lines into mice — revealed
that RPN2 siRNA delivery restored sensitivity to
docetaxel and inhibited tumor growth; these
effects were mediated by the diminished matura-
tion and glycosylation of MDR1 and the accumu-
lation of docetaxel within the orthotopic tumors.
Finally, in a new, albeit small, set of breast cancer
patients, RPN2 expression matched responsive-
ness to docetaxel treatment. — GJC

Nat. Med. 14, 10.1038/nm.l858 (2008).

Focus in

on the perfect job.

We’ve got Careers
down to a Science.

• Job Search

• Resume/CV Database

• Grant Information

• Careers Forum & Advice

• and more...

Science Careers

| from ttx [ourvM Sdtnct KlAAAS [

ScienceCareers.org

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

www.sciencemag.org

Science

1200 Nm York Avenue, N W
Washington, DC 20005

Editorial 202 -326-6550. FAX 202-289-7562
News; 202-3266581 FAX 202-371-9227

Bateman House, 82-88 Hills Road
Cambridge, UK CB2 1LQ

+44 10) 1223 326500, (AX +44 (0) 1223 326501

Susumruok Soviets r or change at address, missing issues, new
orders and renewals, and payment questions; 866434-AAAS (2227)
or 202-326-6417, (AX 202-842-1065. Mailing addresses; AAAS, (>.0.
80* 96178, Washington, DC 20090-6178 or AAAS Member Services.
1200 New York Avenue. NW. Washington. DC 20005
Institutionri Sitr Licenses please call 202-326-6755 tor any
questions or information
Rimini: Author Inquiries 800-635-7181

Commercial Inquiries 803-359-4578
Pcmsuoits 202-326-7074. FAX 202-682-0816
Menhir Butins AAAS/Bames&NoNe.CMn bookstore www.aaas.org/bn ,
AAAS Online Store http;//www.api source.com/aaas/ code MKB6;
AAAS Travels; Betchaci expeditions 800-252-4910; Apple Store
nww.apple/eppstore/aaas; Bank of America MasterCard 1-800-833-
6262 priority code FAA3YU; Cold Spring Harbor Laboratory Press
Publications www.cshlpress.com/affiliales/aaas.htm; GF ICO Auto
Insurance www.geico.com/landingpage/go51.htm7logoel7624;
Hertz 800-654-2200 CDPK343457; Office Depot https;//bsd.
officedepot.com/portaaogin.do; Seabury & Smith Life Insurance 800-
424-9883; Subaru VIP Program 202-326-6417, VIP Moving Services
htipj'/vwrwvipmayllowei.corn.'dciTiestic.indexJhiml; Other Benefits; AAAS
Member Services 202-3266417 crvwwraaaamember.crg.
sdence_editors@aaas.org (for general editorial queries)

sciencejetters@aaas.org (for queries about letters)

5cience_reviews@aaas.org (for reluming manuscript reviews)

sdence_bookrevs@aaas.org (for book review queries)

Published by the American Assodatlon for the Advancement of Science
(AAAS). Science serves its readers as a forum for the presentation and
discussion of important issues related to ihe advancement of science,
including the presentation of minority or conflicting points of view,
raiher than by publishing only material on which a consensus has been
reached. Accordingly, all articles published in Science — including
editorials, news and comment and book review— are signed and reflect
the in df/idual views of iheaudscfs and not offidal points of view adopted
by AAAS or the institutions with which the authors are situated.

AAAS was founded in 1848 and incorporated in 1874. Its mission is to
advance science, engineering, and innovation throughout the world for
the benefit of all people, the goals of Ihe association are toe enhance
communication among scientists, engineers, and the public promote and
defend theintegrity of science and Its use; sirenglhen supper for the
science and technology emetprise; provide a voice for science on societal
issues; promale toe responsible use of science In public policy; strengtoen
and diversify the science and technology workforce: foster education to
science and technology for everyone; increase public engagement with
science and technology; and advance intern atforal cooperation in science.

llttOKMAI ION 1 OK AulHOKS

See pages 634 and 635 of the I February 2008 issue or access

www.sciencemag.org/aboutAiuihors

tonoK-iN-CHirr Bruce Alberts
executive tor ton Monica M. Bradford

DePUTY EDITORS NEWS EDITOR

R. 8 rooks Hanson, Barbara R. Jasny, Colfn Norman
Katrina L Kelner

Eoitormi supervisory si mo* (ono* Phillip 0. Szuromi; senior loiron,
perspectives Lisa D. Chong: senior editors Gilbert ). Chin, Pamela ).
Hines. Paula A. Kiber stis (Boston), Marc S. lavlne (Toronto), Beverly A.
Purnell L Bryan Ray, Guy Riddihough. H. (esse Smith. Valda Vinson;
assoctnif editors )ake S. Yeston. Laura M. Zahn; otuvf editor Stewart
Wills; usouMt mum tufoRs Robert Frederick. Tara S. Marathe: mi
content dcveiopir Martyn Green: boor review editor Sherman ). Surer;
associate i etters Eoitot Jennifer Sills; ediidriai mananir Cara Tate;
senior cDfr EonoRs Jeffrey F. Cook. Cynthia Howe. Harry Jach, Barbara P.
Ordwsy. Trista Wagoner; com odors Chris Filiatreau. Lauren Kmec,
Peter Mooreslde; editoemae coordinators Carolyn Kyle. Beverly Shields:
pint iiAnoes assistants Pamatoulaye Diop. Joi S. Granger. Jeffrey Hearn.
Lisa Johnson. Scon MHIer, Jerry Richardson. Jennifer A. Seibert. Brian
Whtte. Anita Wyrn; eouoriaiauisianis Carlos L Durham. EmltyGulse.
Patricia M. Moore: ixrcunvi assistant Sylvia S. Kihara. administrative
support Maryrose Madrid

News oeputt news eoiidrs Robert Coontz. Elicit Marshall Jeffrey
Mervis. Leslie Roberts; coniribuiinn ioitors Flizabelh Culona, Polly
Shulman; news weiiers Yudhljtl Bhattachar]ee, Adrian Cho. Jennifer
Couztn, David Grimm, Constance Holder. Jocelyn Kaiser, Richard A.
Kerr. Eli Kintisch. Andrew Lawler (New England). Greg Miller.
Elizabeth Pennisi. Rober F. Service (Pacific NW). Erik Stokstad;
interns Rachel Zelkowitz. Andrea Lu. Fayana Richards: coniruuiinn
corresponoenis Jon Cohen (San Diego. CA), Daniel Ferber. Ann
Gibbons. Mitch Leslie, Charles C. Mann. Virginia Morell Evelyn
Strauss, Gary Tiubes; copteoiidrs Linda 8. Felaco. Melvin Gatling;
aominkiraiwe suppori ScherTaine Mack. Fannie Groom; bureaus New
England: 207649-7755. San Diego, CA: 760-942-3252. FAX 760-
942-4979. Pacific Nonhwest: 503-963-1940
Proouciion director James Landry, senior manamr Wendy K. Shank;
assistant manager Rebecca Doshi; senior speciaiists Steve Forrester.
Chris Redwood; sprcuuisr Anthony Rosen; Preeiinmt oiRrcroR David M.
Tompkins; maaanir Marcus Spiegler; sttciaiist Jessie Mudjilaba
Art ooictoRYjel Kats; associate bit MREcioRAar on Morales; iuusiraiors
Chris Bided KatoarineSutlifi; senior ari associates Holly 81 to op. Laura
C reveling. Preston Huey. Nayomi Kevitiyagala; associate Jessica
Newfietd; photo eoitor Leslie Blizard

Science iNtERNAiiONAt

Europe (science@science4nt.co.uk) eddoriai= intcrnauonai manaciDiin
editor Andrew M. Sugden; senior eddorpers wc eeces Julia Fahrenkamp-
Uppenbrink: senior eoiiors Caroline Ash. Stella M. Hartley, tan S.
Osborne. Peter Stern: editoriae support Deborah Dennison. Rachel
Roberts. Alice Whaley; administrative support John Cannell. Janet
Clements, Louise Smith; niwv iuropi news editor )ohn Travis; oeputt
news editor Daniel Clery: contrirutinn corresponoenis Michael Batter
(Paris). John Bohannon (Vienna). Martin Enserink (Amsterdam and
Paris). Gretchen Vogel (Beilin); intern Lauren Cahoon

Asia Japan Office; Asca Corporation. Eiko ishloka. Fusako Tkmura, 1-
8-13. Hirano-dio, Clwio-ku, Osaka-shi, Osaka. 541-0046 Japan; +81
(0) 6 6202 6272. FAX +81 (© 6 6202 6271: asca@os.guH.or.jp: asia
news editor Richard Stone (Beijing: rstone@aaas.org); coniribuiinn
correspondents Dennis Normile (Japan; +81 (0) 3 3391 0630. FAX
+81 (0) 3 5936 3531; dnormile@golcom); Hao Xin (China; +86 (0)
10 6307 4439 or 6307 3676. FAX +86 (0) 10 6307 4358;
cindyhao@gmail.com); PaHava Bagla (South Asia: +91 (0) 11 2271
2896; pbagla@vsnl.com)

Aerka Robert Koenig (contributing correspondent. rcb.ko erf g@grr Acorn)

executive PUBUSKR Alan I. Leshncr
publisher Belli Rosner

FuiniuREin Stsiuesano Operations (m embership@aaas.org); oiRrcroR
Waylon Butler; senior systems anaitst Jonn y Blaker; customer service
supervisor Pat Butler; speciaiists Latoya Casteel Lal/onda Crawford.
Veto Linton; daia entry supervisor C in hi a Johnson; sptuAiistTarnla Hill

Business Operations and Administration director Deborah Rivera-
Wien hold: assistant director, rusiress operations Randy Yi; mananir.
business anrivsis Michael LoBue; mananer. business operaiions
Jessica Tierney; einanciai anaitsts Benjamin Arcnin. Priti Pamnani;
rinhts ano pi emissions; AOMtNisiRAioR Emille David: ASSOCIATE
Elizabelh Sandler; marrfiian director John Meyers: marrftinn
mananir Allison Pritchard; mar retina associates Aimee Aponte,
Alison Chandler, Mary Ellen Crowley. Marcia Leach, Jufianne Wielga.
Wendy Wise: iniirnaiionai marremnn mananir Wendy Sturley:

MARRI DNN EXECU1RJI Jennifer Reeves; MAtCUMNMi MSI A MRVKTSFXKUlief

Linda Rusk; sri ikense saies director Tom Ryan: saies mananir Russ
Edra: saies ano customer service Iquo Edim. Kiki Forsythe.
Catherine Holland. Ilese Ominsky. Phillip Smith. Philip TsoUkidis;
iiitiRONic media: mananer Lizabeth Harman; project mananir Trista
Snyder: assistrnt mananir Lisa Stanford; senior proouciion speciaiists
Christopher Coleman. Walter Jones; proouciion speciaiists Hichele
Johnston, Kimberly Oster
Aoviriiunn om«c tor, woridwrii ao saies Bill Moran
Proouct (sctence_adveitising@aaas.oig); Miomsr Rick Bongiovanni:
330-405-7080. FAX 330-405-7081; west coase/w. canaoa Teola
Young: 650-964-2266; east coasoi. canaoa Laurie Faraday; 508-
747-9395. FAX 617-507-8189: wuiuropiasia Tracy Holmes; +44
(0) 1223 326525. FAX +44 (0) 1223 326532; iapan Mashy
YOshikawa: +81 (0) 3 3235 5961, FAX +81 (0) 3 3235 5852; senior
iraffic associate Defandra Simms
Commircmi Editor Sean Sanders: 202-326-6430
Proiici Director, Ouireacn Brianna Blaser
Ciassimeo (advertfse@sriencecareers.org); us: hecruitmihi sues
■aranir Ian King: 202-326-6528. FAX 202-289-6742; inside saus
mananer : midwe st cakaqa Daryl Andei ion; 202-326-6543; insioe sales
RiPREscuiAtivi Karen Foote; 202-326-6740; ret accourt maranir
Joribah Able: rorihlasi Alexis Fleming: 202-326-6578; southeast
Tina Burks; 202-326-6577; wtsrhticholas Hintibidze: 202-326-6533;
saus cooroirators Erika Foard. Rohan Edmonson. Shirley Young;
RiiiRBAiitHiAiisAmKARANFR Tracy Holmes: +44 (0) 1223 326S25, FAX
+44(0) 1223 326532;saies Marlum Hudda. Alex Palmer. Alexandra
Sorgente; saus assisiant louise Moore; iapad Mashy Yoshikawa: +81
(0) 3 3235 5961. FAX +81 (0) 3 3235 5852; aovertisinn production

OPERATIONS MANANER DebOEah TompklnS; URIOR PROOUCnOR SPECIAIISTS

Robert Buck. Amy Hardcastle; senior iraffic associas Christine Hall.
ruuicATioRs assesiant Mary Lagnaoui

AAAS Board or Directors redrinn presntint, chair David Baltimore;
ERESBiNi James ). McCarthy; prisidinteuci Peter C. Agre miAsu«R
David E. Shaw; chre txicurrvt omicir Alan I. Leshner; norrd lynn W.
Enquisl Susan M. Fitzpatrick. Alice Gasl Linda P. 8. Katehi. Nancy
Knowlton, Cherry A. Murray. Thomas D. Pollard. ThomasA. Woolsey

HI AAAS

ADVANCING SCIENCE, SERVING SOCIETY

SENIOR EDITORIAL BOARD

Kobad M». Unit. of

Marcia McNutt, 4foi
Linda Partridge, Un
Vera C. Rubin, Can

BOARD OF REVIEWING EDITORS

Didd Altshuler. Brood imftftj fr

Angelika Amon, Mil

McinraE 0. Andrea e, Max Hand Imt. Manx
Kristi S. Aiuath, to. of Colorado
John A. Baruh, Vote Uni*

fc™BiwTO.Sf , i^53lr^SfeoJ h ^

VLUxtgzmtestr

Michael J. Sevan, linn, of Waiting* on

Ko^^?l?*X'‘«iondUt

Peer Burk. (MSI

Paul M. Brakefieid, laden Uni v.

nlSdiool

Joseph A. Bums. Cornell Unit.

Will uni P. Bpti. Population Rrfrrrnrr Bureau
Peter CanetTiet. Uni * oflru/rn VIS

Ger brand Coder. AW

SiridrciapSan^Sitorn^i^Hovfai Bo it on
“ Clary, Oxford Untvenily

Stephen M. Cohen, lematet lie Sdenm lab. Singapore
William Cumberland, Unrv. of California, lor Angela

£E 8o,,on

Enwean auU O wmitiakii. Welcome butt Sanger tin.
Dennis Oischnr, tjr.r*. ofPermyl/rmk

Pn2 )SSSr,i^&cSf^7r& ,mt

W. Ford Doolittle, Dakota ie Utie.

Jennifer A. Ooudna, Uni/, of California. Berkeley

r.l i i. - II,, ~

Erhard lrttv.-Vn , fWSp/-/n.riruC Serin
Oouglas H. Erwin, SniThiorann Initiation

PaoiG. Falkowslri. Bufgen On*.
W^^penhagen

Alain Fischer, MM
Stott E. Fraser. CalTech
Chris D. Frith, Un* College London
Wullrain Gerstner, IPR latnanne

&n - .“G Sc

nnis l. Hartmann, bhv. of Waiting
--.-■11 H-mkrce/ortll i

Marlin Heimann. dai Monti lmt„ fen a
James A. Hendler, Bennetaer Pat/tedmk. hal

BSsaassB*—

sjmesieviifiSK. «

Stephen Jackson, Un/. of Cambridge
Stiven Jacobsen. Uni/, of Cob forma lot Angela
Peter Jonas, Um/enBdl feeburg

Barbara B. Kahn, Harvard Medical School

^yTo^Unr/. Cd^AS
Bernhard Kviieer. Max Planck In d. Stutlgmt
Elisabeth A. Keilog, Unit. ofMtnouri SUotm
Alan_B. Krueger^Pnnrrfon Urtrr.

Kump f

01 Ic UndvaU, Um* Harped, land
iScSa^uSSu Stare ard Uni*.

Kelo. C.ntrioeXrod. otScmica
Andrew P. MacKeiuie. Uni/. ofStAndrem
Raul Madariaga, f tote Notrmdr Supaieure. Peril
Anne Magurran, Uni/, of Sf Andrew)

Michael Aalim, ttng'i College, toed,

Urbano- Champaign Thomas Stocker,

Georg SchuU, Albert l udnigi- Uni/en ft dr

-I-'m'. 'j'i ' ! : ! , i I ’

Ktrrence J . Seinowslri, Ihe Salkloilifulr
fcTOg^m ^ sK &ltiiilfcabfortia Berkeley

i“Jbefh'siwn?HH7uTHh

~ '“odtRf, Uni* of Bern

■may Virginia Commomurdth Uni/.
ing, Un* o ikmiutly
er 'Lorigne, Cenentech
pp. Unit, of latnanne

— - - n det Klis. Arironomicd In it ofAmilerdam
Derek van dvr Kooy. i/nhr. ojhronio
Bert Vogetslvin, John) Hopkim Uni/.

Ulrich M. von Andrian. Harvard Medical Schod
Christopher A. Walsh. Harvard Medical School
Graham Warren, idle Itirv. School of Med.

Melga Nowolny. Furtprar. nr

Elinor Ortrom, Indiana Uni*.

Philippe Poulin, CJV»

Mary Poerer. Unit, of California. Sake.

SSrs: acttSSBr

Lvs Real, imoty Uni/.

Colin Kenfrew, Unit, of Cambridge

KaTlura Konianowka OmtdCaafomia. Berkeley

John k. Vain III. The Saippi Bn. tin.
Jan Zaancn. teiden Un i*

Martin ZaRs, NIMH, NIH

Huda teyhbi, Bo/loc College of Medicine

BOOK REVIEW BOARD

John Aldrich, Duke Uni/.

Da rid Bloom. Harvard Univ.

Angela Creager, Princelon Uni *

Lewis Wolpert, Urn*. College London

Atinc^betk Brseaixh

1274

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sdencemag.org

EDITED BY CONSTANCE HOLDEN

Cane toad races area barroom pastime in parts
of northern Australia. Now the pestiferous
amphibians are racing for science.

Researchers led by zoologist Michael
Kearney at the University of Melbourne have
been trying to predict the potential range of
the country's plague of cane toads, which were
first introduced in the 1930s to attack sugar
cane beetles in Queensland.

The toads have been expanding their range
roughly at a rate of 60 kilometers a year. To see
how far south the population could extend, the
scientists tried to gauge how far the animals

can hop under vari-
ous temperature
conditions. They
tested 89 toads on a
2-meter course,
measuring hopping
speed at five tem-
peratures ranging
from 15° to 35°C.
Hopping speeds
ranged from a
molasses-like 300
meters per hour at
15°C to a brisk 2.2
km at 30°C, they
report in the August
issue of Geography .

Combining data

on toad movements with information on
reproductive needs (ponds for eggs and lar-
vae) and climate, the researchers predicted
that, contrary to some previous analyses, the
toads won't be invading major cities such as
Sydney and Melbourne. Even with predicted
climate change, they say, adults would be too
slowed down by cool, dry weather to spawn or
find enough to eat. Biologist A. Marm Kilpatrick
of the University of California, Santa Cruz, who
does research on the West Nile virus, calls it a
"neat' study that "offers a bottom-up mecha-
nistic way to look at an animal's distribution'
by combining data on climate, physiology,
and behavior.

a

2

&

i

a

The Emperor's Toes

This 80-centimeter-long foot — in what archae-
ologists describe as "exquisitely carved army
boots covered with a
lion skin and deco-
rated with tendrils
and Amazon
shields" — is part of a
5-meter-tall statue of
the Roman emperor
Marcus Aurelius, who
ruled from 161 to
180 C.E. The frag-
ments were found last
month in the rubble
of a huge bath com-
plex in the Turkish
town of Sagalassos by
a team led by Marc
Waelkens of the
Catholic University of
Leuven, Belgium. The

baths were destroyed by an earthquake about
600 C.E., according to carbon dating of owl
pellets at the site.

The statue was part of a gallery of 2nd cen-
tury emperors that the scientists
believe stood in niches around the
cross-shaped, mosaic-covered
frigidarium, into which people
plunged for cold baths. Last year,
the team uncovered chunks from a
giant statue of Hadrian, who ruled
from 117 to 138 C.E., that are cur-
rently on display in the British
Museum in London.

The Right Note

Perfect pitch is thought to be a rare
capacity, possessed by about one in
10,000 people. But researchers at
the University of Rochester's
Eastman School of Music in New
York state have developed a way to

test nonmusicians for perfect pitch that they
hope will yield a more accurate estimate.

The usual test — playing a note and seeing if a
person can identify it by the sound — can be done
only with subjects who know musical notation.
Betsy Marvin, a musical theorist, and Elissa
Newport, a neuroscientist, devised a test in which
people listen to a three-note motif played repeat-
edly for 20 minutes. Then they hear either those
notes again or the same motif transposed to a dif-
ferent key and are asked to identify the original
notes. The researchers first validated the test with
music students, comparing results with the results
of the traditional pitch test. Then they tried it on
24 volunteers with little or no musical training. Six
of those proved to be as accurate or almost as
accurate in recognizing pitches as the music stu-
dents with perfect pitch. That suggests perfect
pitch is more common than has been thought,
says Marvin, who presented preliminary results
last week at the International Conference of Music
Perception and Cognition in Sapporo, Japan.

Peter Gregersen, a geneticist at the North
Shore-Long Island Jewish Health System in New
York who studies absolute pitch, says such a test
could help in determining if people are born
with perfect pitch or if learning plays a role.

Chinese Emissions

Economic modelers are having a hard time keep-
ing up with the Chinese industrial juggernaut.
Based on recalibrated modeling and the latest
economic data, a new working paper from the
Electric Power Research Institute (EPRI) in Palo
Alto, California, forecasts that by 2030 China's

energy sector will be emitting about 4 billion tons
of carbon — twice as much as estimated in a 2005
International Energy Agency (I EA) projection.
"Growth in China is so rapid that it is difficult to
predict emissions just 2 years from now," says the
unpublished paper by Geoffrey], Blanford and
colleagues at EPRI and the Centre for Energy
Policy and Economics in Zurich, Switzerland.

www.sciencema 9 .org SCIENCE VOL 321 5 SEPTEMBER 2008

1275

EDITED BY YUDHIJIT BHATTACHARJEE

LHC

The world’s most expensive
particle physics experiment
will get imderway on 10
September when CERN’s
Large Hadron Collider
(LHC) goes online (see p.
1291). Thousands will be
watching as the first beams
of protons are sent through
the collider. But the search
for the Higgs boson also
has a human side. Here are
a few of their stories.

So, on the big day, who will be sitting in the
Captain Kirk chair ready to push the start but-
ton? Roger Bailey, head of the beam commis-
sioning team, isn't let-
ting on. "It'll be me or
one or two col-
leagues," he says.

"We'll decide over the
next couple of days."

Bailey has assem-
bled a 30-person
team that will operate
the accelerator round
the clock over the
next few months in
hopes of ironing out the kinks that come
with the first collisions, "it's going to be
pretty stressful between now and Christmas,"
he says.

Bailey was there in the control room in
1989 at the opening bell for CERN's previous
big machine, the Large Electron-Positron col-

lider. "Half the lab was crammed in there,"
he recalls. His goal for 10 September is
clear, if not simple: Get streams of protons
to circumnavigate both rings. "If we can do
that, we'll know we have something we can
work with."

If the Large Hadron Collider (LHC)
reveals the Higgs boson — the famed God
particle — then CERN officials have a
detailed protocol to announce the discov-
ery at a specially convened seminar.
However, CERN spokesperson James
Gillies concedes that word will probably
leak out prematurely. "People are excited
about (LHC), and they want to talk,” he
says. "I think it's pretty likely that if there's
solid evidence of the Higgs, it will come
out." Still, Tejinder "Jim" Virdee, leader of
the team working with the massive CMS
particle detector, says that he expects his
collaborators — all 2900 of them — "to fol-
low the protocol, no ifs or buts."

^THEY SAID IT

"Science, science, science, and science."

— House Speaker Nancy Pelosi (D-CA), at a 26
August breakfast meeting during the Democratic
National Convention in Denver, when asked
about her plans for the first 100 days if Barack
Obama is elected president. A spokesperson says
the speaker’s 'innovation agenda" also includes
l '21st century jobs and biomedical research."

In Other News ...

AWARDS

Three European scientists last week received
honors for their efforts to communicate sci-
ence to the public. Chris Smith, a virology lec-
I turer at the University of Cambridge, U.K.,
j received the Royal Society's $4500 Kohn
j award for his popular radio show, The Naked
► Scientists, which aims to "strip science down to
3 its bare essentials" through interviews with
? researchers. Evolutionary biologist Axel
I Mever of the University of Konstanz. Germany.

took home $7300 as winner of the European

Molecular Biology Organization's (EMBO's) annual communications award for his articles in main-
stream newspapers and magazines and his radio and television appearances. EMBO also awarded
a special prize of $4700 to another successful science communicator: zoologist Jurgen Tautz of the
University of Wurzburg, Germany, who authored a popular book on honey bees last year.

Q&A

Three months after the Large Hadron
Collider (LHC) is turned on, former
CERN particle physicist Rolf-Dieter
I leu or will rejoin
the lab as its new
director general.
Currently research
director for parti-
cle and astroparticle
physics at DESY,
Germany’s parti-
cle physics lab in
Hamburg, Heuer
says that it’s time for CERN “to change
back to analysis mode” after spend-
ing more than a decade building the
S5.5 billion machine.

Q: CERN still has debts to repay on the
LHC. What will be their impact?

[Repayments] will certainly limit our activi-
ties, but after 2010-11 we will have some
maneuvering space to fund new initiatives.

Q: Will these include non-LHC areas,
such as antimatter and neutrino
physics, that were scaled down during
construction?

These have continued on a minimal level,
but we do need to maintain diversity. If I
want to ramp these up, proposals must be
scientifically high-class. One should be
able to convince funding agencies.

Q: Funders seem cool about pushing
ahead with the International Linear
Collider. Would a delay give CERN's
emerging CLIC accelerator technology
a new opportunity over the design
already on the table?

Between 2010 and 2012, the LHC will tell
us the next energy range of interest. Once
we know the energy range, we can decide
on which technology.

0: Should countries such as the United
States and Japan, which don't pay for
the collider or for operating costs,
start paying a share of CERN's annual
running expenses?

There has to be a discussion of the role of
CERN as a European lab in a global part-
nership, and the issue of contributions will
come up naturally. We must be proactive:
Already next year, we should start to
define how that partnership would look.

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1277

GENETIC PRIVACY

Whole-Genome Data Not Anonymous,
Challenging Assumptions

Last week, scientists learned that a type of
genetic data that is widely shared and often
posted online can be traced back to individuals
who proffered up their DNA for research The
revelation, in a paper published in PLoS
Genetics, prompted the National Institutes of
Health (NIH) in Bethesda, Maryland, and the
WellcomeTrust in the United Kingdom to strip
some genetic data from their publicly accessi-
ble Web sites, and NIH to recommend that
other institutions do die same.

The concern is with studies in which
researchers pool genetic data from hundreds of
people to look for broad patterns of
genetic inheritance. Because the pool
consists of DNA from so many people,
the assumption has been that it would be
impossible to identify any one individ-
ual’s DNA. The new study suggests
that’s not the case. NIH officials and
others agree that the likelihood of a
breach of privacy is low, laigely because
the pooled data must be matched
against a particular person s isolated
DNA — something that, currently, only
researchers generally have access to.

But the discovery' that these DNA
pools don’t protect anonymity is still
troubling, especially because no one
had considered that a possibility.

The first response to the results
“is, ‘You’re crazy,’ ” says David
Craig, a geneticist at the Translational
Genomics Research Institute in
Phoenix, Arizona, who conducted the
work. Less than 9 months ago, NIH
was so confident in the anonymity of
pooled genetic data that it recom-
mended it be made public for all
researchers to use.

Craig found this confidence misplaced, for
a simple reason: Geneticists now routinely
examine hundreds of thousands of DNA vari-
ants, called single-nucleotide polymorphisms
(SNPs), at a time, instead of hundreds as they
did just a few years ago. As a result, they’re
gathering enough information about the pat-
tern of SNPs in a pooled sample that it’s feasi-

ble to deduce whether a particular individual,
with her own unique SNP blueprint, is repre-
sented in a much bigger pool of DNA — even if
that person’s DNA was less than 1% of the
mix. Craig and his colleagues managed to do
this by ascertaining the distribution pattern of
every single SNP — essentially, asking the
same question 500,000 times. They were suc-
cessful because, it turns out, every individ-
ual shifts a genetic pool subtly in certain
directions, and studying enough SNPs
unveils the pattern of those shifts. The
biggest chance of error comes from false posi-

tives from relatives whose DNA may also
appear in the pool, says Craig.

NIH oft' cials were startled when Craig noti-
fied them of his findings about 2 months ago;
they had their own statisticians repeat the exper-
iments. “They said, ‘Yup, this works,’ ” says
Elizabeth Nabel, head of NIH s Heart, Lung,
and Blood Institute. “We still consider the risk
to the individual relatively low,” she continues.

but “there’s a window of vulnerability.”

The greatest concern is that identifying an
individual this way could reveal sensitive health
information. Genome-wide association studies
compare data from people with and without a
particular disease, so knowing which pool a
person falls into can convey whether they have,
say, cancer, or diabetes, or multiple sclerosis.
“We have a false sense of security with pooled
data,” says Pablo Gejman, a psychiatric geneti-
cist at Northwestern University in Evanston,
Illinois. “There is sensitive information” here.

The Wellcome Trust has pulled data on
about a dozen common diseases, and NIH
has pulled data from nine genetic studies ofT
two sites, dbGaP, which includes genome-
wide association studies, and CGEMS, a site
for cancer genetics work. The seven affected
studies on dbGaP had been downloaded by
about 1000 people all told, says James Ostell,
who oversees that and other NIH databases.

NIH officials are informing geneticists
about the policy change through e-mails and
their Web site; the Broad Institute in Cam-
bridge. Massachusetts, has followed suit
and removed pooled data from its site. This
is “a logical choice, a necessary choice,” says
Michael Boehnke, a statistical geneticist at
the University of Michigan, Ann Arbor,
whose data from a diabetes study was taken
down from NIH.

Nabel says that NIH is considering a new
policy in which the pooled data will be
released to researchers who apply, as is now
the case with data traditionally considered
much more sensitive.

Still, Ostell and others say the current pri-
vacy risk is minimal. It could be of more con-
cern 5 or 10 years from now, as genetic infor-
mation proliferates. One possible scenario is
that law- enforcement agencies might turn to
pooled data to determine whether their suspect
is present — and even demand that the
researcher help them identify him.

Craig’s work could help future forensic
investigators in another way: Currently, they’re
unable to identify a suspect’s DNA in a mixed
sample — say, a sample of blood from several
people — if the suspect's blood is less than 1 0%
of the total. “A lot of forensic crime samples do
have small contributions from people of inter-
est, [and) right now we can do essentially' noth-
ing,” says Bruce Weir, a biostatistician who
studies genetics and forensics at the University
ofWashington, Seattle. -Jennifer couzin

i

t

1

2

1

e

§

1278

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

PLANT SCIENCE

China Plans $3.5 Billion GM Crops Initiative

BEIJING — Confronted with land degradation,
chronic water shortages, and a growing popu-
lation that already numbers 1.3 billion, China
is looking to a transgenic green revolution to
secure its food supply. Later this month, the
government is expected to roll out a $3.5 bil-
lion research and development (R&D) initia-
tive on genetically modified (GM) plants.
“The new initiative will spur commercializa-
tion of GM varieties,” says Xue Dayuan, chief
scientist on biodiversity at the Nanjing Insti-
tute of Environmental Science of the Ministry
of Environmental Protection.

A central aim is to help China catch up with
the West in the race to identify and patent plant
genes "of great value,” says Huang Dafang.
former director of the Biotechnology Research
Institute of the Chinese Academy of Agricul-
tural Sciences in Beijing. Once intellectual
property rights are in place, says Huang, trans-
genic technology could transform Chinese
farming “from high-input and extensive culti-
vation to high-tech and intensive cultivation”

In the decade since China first allowed
commercial planting of four GM crops, the
government has moved cautiously, granting
only two further approvals for small-market
species: poplar trees and papaya (see table).
Currently, just one GM crop — insect-resistant
cotton — is now planted widely, says Xue.
China has balked at commercializing
GM versions of staples such as rice, corn,
and soybeans.

That may change, as China s leadership has
thrown its weight fully behind GM. "To solve
the food problem, we have to rely on big sci-
ence and technology measures, rely on
biotechnology, rely on GM ,” Premier Wen
Jiabao told academicians last June at the
annual gathering of the Chinese Academy of
Sciences (CAS) and the Chinese Academy of
Engineering. China’s State Council, which
Wen leads, approved the GM initiative in July.

Details of the new initiative, including
a which crops will gain initial support, are being
| hammered out, scientists say. Some funds will
§ go to R&D on transgenic 1 ivestock, an area that
£ has lagged behind GM crops. By 2006, the
g Chinese government had granted permits for
P 21 1 field trials of 20 GM crops, including the
£ six approved for commercial production. As in

other countries, the varieties that China has
commercialized so far are equipped with genes
to resist pests, tolerate herbicides, or stay fresh
longer — not genes that directly boost yields.

Proponents note that China’s cautious

embrace of transgenic technology has yielded
a major success story: GM cotton. Introduced
into commerce in 1997, 64 varieties of pest-
resistant cotton are now grown on 3.7 million
hectares, or about 70% of the area devoted to
commercial cotton, averting the use of
650,000 tons of pesticides, says Huang.

The big prize is GM rice. Three years ago,
Huang Jikun, director of CAS’s Center for Chi-
nese Agricultural Policy in Beijing, and col-
leagues reported that f eld trials of GM rice in
China were going well — boosting yields and
reducing pesticide use on plots — and predicted
that the varieties were on the threshold of com-
mercialization (Science, 29 April 2005, p. 688).
But the Chinese government is reluctant to tin-
ker with the country's most important crop and
has put off commercialization. The new initia-
tive might break the logjam, says Huang Jikun.

“I hope the commercialization of GM rice will
come within a couple of years,” he says.

Although the central government has not
released a budget f gure for the new initiative,
a spokesperson for the Ministry of Agriculture
told Science that it would cost
$3.5 billion over 13 years. Half is
expected to come from local gov-
ernments on whose land GM
crops will be grown and from
agricultural biotechnology com-
panies. “It’s a new way to support
a big science project in China,”
says Huang Dafang. Another
departure from other R&D initia-
tives, he says, is that each funded
program is expected to produce an
economic payoff.

One component of the initiative
will be to educate the public about
GM crops, says Huang Jikun.
Although China is unlikely to see
the sort of protests that have
derailed field trials and commer-
cialization in Europe, there are cur-
rents of disquiet in the general pop-
ulation. “For consumers, the safety
of GM crops is the biggest worry.
Just like some people are afraid of
ghosts, some people are afraid of
GM crops,” says Zeng Yawen of
the Biotechnology and Genetic
Resources Institute of the Yunnan Academy of
Agricultural Sciences in Kunming. Although
Zeng believes that GM food safety will be
demonstrated adequately, he worries that the
new initiative will push China to “move too fast
to commercialize GM varieties.”

But with questions mounting about China’s
ability to feed itself, others contend that not
pushing ahead with GM varieties could be
more detrimental than any theoretical hazard.
“Any kind of new technology may have risk,”
says Huang Dafang. But legitimate concerns,
he says, should not be overshadowed by scare
tactics designed to “mislead the public in the
name of environmental protection.” With the
country’s leaders firmly behind GM crops, it’s
unlikely that any protests would get very far.

-RICHARD STONE
With reporting by Chen Xi and ]ia Hepeng.

CHINA'S TRANSGENIC PLANTS

"* .«r

f i 9 'S r ‘

-Cotton

r

*

Stoeet Pepper

p$i

[Papaya^

PLANT

YEAR COMMERCIALIZED

Cotton

1997

Petunia

1997

Tomato

1998

Sweet pepper

1998

Poplar trees

2005

Papaya

2006

Slim pickings. Of the six plants that China has approved for com-
mercialization, only cotton is grown widely. A new initiative could
pave the way for GM versions of the biggest prize of all: rice.

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1279

NEWS OF THE WEEK

CANCER GENETICS

A Detailed Genetic Portrait of the
Deadliest Human Cancers

Three studies published this week have
given researchers their most detailed look
so far at the genetic mutations that underlie
the deadliest of human cancers: pancreatic
cancer and the brain tumor glioblastoma.
They have firmed up the role of key genes
and also found that scores of aberrant genes
are involved in relatively few cell signaling
pathways. One study also unearthed a gene
never before linked to cancer that is
mutated in a substantial fraction of glioblas-
toma tumors. “It shows we can still be sur-
prised” by the biology of cancer, says
Michael Stratton, who oversees a cancer
gene sequencing project at the Sanger Insti-
tute in Hinxton, U.K.

These studies are all based on the prem-
ise that information gleaned from systemat-
ically cataloging the main mutations in
tumors will be worth the high cost. Three
years ago, when genome sequencer Eric
Lander of the Broad Institute in Cambridge,
Massachusetts, proposed spending SI. 5 bil-

lion on what is now called The Cancer
Genome Atlas (TCGA), skeptics helped
persuade the U.S. National Institutes of
Health to start with a 3-year, $100 million
pilot project. One of the glioblastoma stud-
ies is the first fruit of that effort.

Meanwhile, a team led by Bert Vogelstein,
Kenneth Kinzler, and Victor Velculescu at
Johns Hopkins University in Baltimore,
Maryland, had begun a private cancer
genome project, starting with breast and
colorectal cancer ( Science , 8 September
2006, p. 1370). Now this team and collabora-
tors have sequenced the coding regions of
20,700 genes — nearly all the known genes
in the human genome — in 22 glioblas-
toma and 24 pancreatic cancer samples.
They also looked for abnormalities in gene
copy number and gene expression.

In two papers published online by
Science this week (www.sciencemag.
org/cgi/content/abstract/1 164382 and
-1 164368), they report finding hundreds of

genes that were mutated in these two can-
cers. There were an average of 63 altered
genes in each pancreatic tumor and 60 per
glioblastoma. The mutations varied from
tumor to tumor, but the most important
tended to fall in the same cell pathways. For
example, 12 specific pathways were dis-
rupted in at least 70% of pancreatic tumors.
“It points to a new way of looking at cancer,”
says Vogelstein, who suggests that treat-
ments should target these pathways, not the
products of single genes.

One of the altered genes found in the
glioblastoma study, IDH1, appeared in 12%
of tumors, and more often in younger
patients and those with secondary tumors,
the Johns Hopkins team reported. A change
in an amino acid of the encoded protein
seems to help patients with this mutation
live longer than others with glioblastoma.

The third study, published online by
Nature, analyzed more than 200 glio-
blastoma samples. It surveyed all the
samples for genetic alterations such as
changes in copy number and probed about
half the samples for mutations in 600 genes
already implicated in cancer, says co-leader
Lynda Chin of the Dana-Farber Cancer
Institute in Boston ( Science , 4 July, p. 26).
The study found many of the same aberrant

Memory Recall

and left arms each time they ran the maze. In
between runs, the rats spent 1 0 to 20 seconds
on a running wheel.

During this delay period, neurons in the
hippocampus fired in sequences that pre- g
dieted which arm the rat would run next, the z
researchers report on page 1322. Even in the *
few cases when a rat goofed and went the f
wrong way, the preceding firing sequence |
predicted its mistake. These sequences — |
which resemble sequences that occur as a |
rat actually runs through a maze — likely s
represent the brain’s internal mechanism for 8
planning (or reminding itself) what it has to 2
do next. Buzsaki says.

The findings confirm a decade-old pre- |
diction that the hippocampus might generate |
such firing sequences to maintain important 3
information during a delay in a task, says ^
David Redish, a neuroscientist at the Univer- |
sity of Minnesota, Minneapolis. Redish ^
notes that consistent patterns of activity §
emerged only when the rat had something to c
remember. “When the rat is just running on 8

NEUROSCIENCE

Hippocampal Firing Patterns Linked to

Memory aid. A rat's hippocampus (above) generates sequences of
neural firing that may help it remember what to do next.

The hippocampus, tucked deep
inside the temporal lobes of the
brain, has been intensely studied
for its role in recording memo-
ries. Now two studies — one with
rats and one with people under-
going surgery for intractable
epilepsy — suggest that patterns
of neuron firing in the hippo-
campus are also involved in
recalling past experiences.

“The two papers are signifi-
cant because they point directly
to reactivation of neural activ-
ity sequences as a mechanism
for memory recall,” says Edvard Moser, a
neuroscientist at the Norwegian University
of Science and Technology in Trondheim.
Such a mechanism may underlie several
functions attributed to the hippocampus,
Moser says, including navigation, memory;
and planning future actions.

In the rat study, researchers led by Eva
Pastalkova and Gyorgy Buzsaki of Rutgers

University' in Newark, New Jersey, simulta-
neously recorded the activity of scores of
hippocampal neurons as rodents ran through
a maze shaped like a squared-off figure
eight. The rats always started the maze by
running down the middle of the three arms
and then chose to continue down either the
left or the right arm. The researchers
trained them to alternate between the right

1280

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

a wheel for the heck of it in its home cage,
they don’t see it.”

In the human study, published online
this week in Science (www.sciencemag.
org/cg i/content/abstract/ 1 164685),
researchers led by Hagar Gelbard-Sagiv of the
Weizmann Institute of Science in Rehovot,
Israel, and Itzhak Fried of the University of
California, Los Angeles, recorded from hun-
d dreds of neurons in and around the hippo-
campus of 13 epilepsy patients undergoing
S operations in which smgeons introduced elec-
§ trades into the brain to locate the source of
I their seizures. The patients watched several
§ 5- to 10-second video clips that depicted a
S variety of landmarks, people, and animals. A
8 few minutes later, the researchers asked the
y patients to freely recall the clips they’d just
| seen and call them out as they came to mind,
s (Most subjects easily remembered almost all
| of the clips.) The f rst time the patients saw the
I clips, many neurons in the hippocampus and
§ a nearby region, the entorhinal cortex,
^ responded strongly to certain cl ips and weakly
£ to others— preferring a clip from The Simp-
* sons, say, to ones showing Elvis or Michael
| Jordan. Later, each neuron began firing
u strongly a second or two before the subject

reported recal ling that neuron’s preferred clip,
but not when the subject recalled another clip.

“Previous work [with animals] has shown
that such reactivation occurs during sleep as
well as during certain behaviors where mem-
ory is needed, but it has remained unclear
whether reactivation actually reflects recall
of the memory,” say Moser. Fried’s findings
are exciting because they provide the first
direct link between reactivation of hippo-
campal neurons and conscious recall of a
past experience, says neuroscientist Matthew
Wilson of the Massachusetts Institute of
Technology in Cambridge.

Both studies have implications for an on-
going debate about the relationships among var-
ious functions attributed to the hippocampus,
says Lynn Nadel, a neuroscientist at die Univer-
sity of Arizona in Tucson. Nadel says that the
findings f t with his view that the neural mecha-
nisms underlying spatial navigation, episodic
memory, and action planning may be one and
the same. “One might say at this point that the
available data suggest that the hippocampus is
critical for ‘navigating’ through space not only in
the present but also in the past, to retrieve mem-
ories, and in the future, to predict the results of
actions,” Nadel says. -GREG MILLER

Japanese Budget Rollout

TOKYO — Japan's education ministry last week
optimistically called for boosting fiscal 2009
science spending a hefty 13.4% year-on-year
to S24.1 billion. The ministry wants to add
$20 million, a 12.4% increase, for academic
research grants and 11% more — for a total of
$1.2 billion — to advance big science projects,
including $41 million for Japan's contribution
to the international Atacama Large Millimeter
Array in Chile. Applied research fared even
better. The ministry wants to increase one such
grant category, for example, by 42%, to $4.5
million. The proposed budget faces scrutiny
from the budget-minded finance ministry.
"Negotiations will be tough, but we'll do our
best," says Shinichiro Izumi of the education
ministry. The budget, which takes effect in
April, will be finalized by January.

-DENNIS NORMILE

Taleyarkhan Weighs Suit

Rusi Taleyarkhan, the Purdue University
nuclear engineer deemed guilty of research
misconduct, isn't going quietly. Last week,
Purdue stripped him of his named professor-
ship. Now, Taleyarkhan and his attorney are
considering filing a grievance with Purdue, a
lawsuit against the school, or both. "The
process and the manner in which Purdue has
carried itself ... is testimony for the need to
resort to the court system," Taleyarkhan wrote
in an email to Science. In 2002, Taleyarkhan
and colleagues reported that a tabletop device
generated nuclear fusion inside collapsing
bubbles. But in July, an investigation organ-
ized by Purdue concluded that later reports
aimed at replicating the work involved
research misconduct. Taleyarkhan's attorney
says the scientist will continue to investigate
bubble fusion. -ROBERT SERVICE

Your Local Library

The U.S. National Institutes of Health (NIH)
has chosen nine screening centers in the sec-
ond phase of its Molecular Libraries program
( Science , 8 August, p. 764). NIH wants to test
biological assays submitted by researchers
against 300,000 chemicals in hopes of finding
research probes and drug leads. Four major
centers will receive a total of $208 million over
4 years — the Burnham Institute for Medical
Research and The Scripps Research Institute,
both in San Diego, California; NIH's intramural
center in Rockville, Maryland; and the Broad
Institute in Cambridge, Massachusetts. NIH will
also support five smaller centers.

-JOCELYN KAISER

Probing a killer. Two new studies tally
genetic glitches that cause the brain
tumor known as glioblastoma, orange
in this image of brain celts.

genes that the Johns Hopkins
team uncovered — but not IDHI,
which was not among the genes
the team sequenced. Their larger
sample set will serve as a reliable
reference on how frequently
mutations occur in glioblastoma,
including several genes for
which the evidence was limited
until now, says Chin. Having
methylation data and samples
from patients who received treat-
ment also allowed the team to
finger mutations in DNA repair
genes that may help explain why
tumors that initially respond to temozolo-
mide, the main drug for glioblastoma, can
become resistant to subsequent therapies.

TCGA is preparing follow-on papers, for
example on using the molecular data to clas-
sify subsets of tumors, Chin notes. It will
also expand the search: The project, which is
also studying lung and ovarian cancers, will
use new technologies to sequence thousands
of genes in each tumor.

“I see them [the public and private
glioblastoma studies] as wonderfully comple-
mentary,” says pathologist Paul Mischel of
the University of California, Los Angeles,
who studies glioblastoma. Other researchers
who hope to use the findings to improve
cancer treatment agree. “This is a start and a
wonderful start,” says Santosh Kesari, a
neurooncologist at Dana-Farber.

-JOCELYN KAISER

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1281

MEETINGBRIEFS»

MATHFEST 2008 | 31 JULY-2 AUGUST | MADISON, WISCONSIN

Shapeshifting Made Easy

Mathematicians aren't squeamish about
doing dissections, but they do often come
unhinged. Now computational geometers at
the Massachusetts Institute of Technology
(MIT) in Cambridge have proven it’s possible
to do mathematical dissections without
falling to pieces.

The victims in this case are not frogs but
polygons: simple geometric shapes bounded
by straight sides. In the early 19th century,
mathematicians proved that any two polygons
with the same area can be cut into a finite
number of matching pieces. For example, it’s
possible to cut a square into four pieces and
rearrange them into an equilateral triangle.

About 100 years ago, the English mathe-
matician and puzzle designer Henry Dudeney
added an extra wrinkle to the dissection chal-
lenge: He showed that the rearrangement from
square to equilateral triangle can be done with
pieces connected by hinges (see figure,
above). Dissection enthusiasts have since
devised many more hinged transformations.

In 1997, Greg Frederickson, a computer
scientist and geometric-dissection buff at
Purdue University in West Lafayette, Indiana,
asked whether what Dudeney did for the
square and triangle can be done for any two
polygons. The question caught the attention
of Erik Demaine, then beginning graduate
work in computer science. A decade later,
Demaine, now a professor at MIT, has the
answer: in a word, yes.

Demaine returned to Frederickson s prob-
lem last fall with his father, Martin Demaine,

and four students in a problem-solving semi-
nar: Timothy Abbott of MIT, Zachary Abel
and Scott Kominers of Harvard University,
and David Charlton of Boston University. The
group came up with a general procedure for
turning an arbitrary dissection into a hinged
dissectioa Demaine described their proof at
MathFest. "It was a surprising result to me,
because I thought it was false,” he says.

Their proof starts with an idea “so crazy
that we never thought of it,” Demaine says.
That idea is simply to take an unhinged dis-
section of one polygon and arbitrarily add
hinges, then subdivide the pieces and add
additional hinges until the polygon can con-
tort into its equal-area partner. The key step
is to show that judicious subdivision can, in
effect, take a hinge that connects, say, piece
A to piece B and move it to connect A to C
(see figure).

“The movement is magical,” Frederickson
says. On the other hand, he notes, “you don’t
get very pretty dissections this way.”

The construction works on three-
dimensional (3D) dissections as well, which
could help guide the design of reconfigurable
robots — modular machines that rearrange
their parts like real-life Transformers. In 3D,
unfortunately, equal volume doesn’t guaran-
tee the existence of a dissection. But when
dissections do exist, the MIT group’s con-
struction shows that they can be refined into
hinged dissections. The results are an encour-
aging first step toward applications, Demaine
says: “Now the optimization begins.”

4 Location, location, location. A square can
| become an equilateral triangle without ever falling
apart (top). The same is true for other pairs of poly
gons. The proof starts with a trick that, in effect,
moves hinges around [bottom ).

Sweet Inspiration

Geometers find ideas everywhere. Take
Mozartkugel, the famously spherical choco-
late confections from Austria. Erik Demaine,
his father, Martin, and colleagues John
Iacono at the Polytechnic Institute of New
York University and Stefan Langerman at
University Libre de Bruxelles have worked
out a more efficient way to wrap them.

As mapmakers know from trying to go the
other way, flattening a globe
invariably distorts areas on
its surface. Conversely,
wrapping a globe with an ,
inflexible wrapper (such
as foil) crinkles the wrap-
per with infinitely . —

many tiny folds. As a *8jS
result, the area of any "
wrapper must exceed the ' .
surface area of the choco- ^
late ball [4n square units for |
a ball with a radius of 1).

One popular brand of
Mozartkugel comes in a square
foil of side length nV2 (;i times
the square root of 2). Another comes in a n x
2 n rectangular wrapper. In each case, the
wrapper’s area is 2n 2 — some 57% greater
than the surface area of the sphere. Demaine
and crew set out to see if they could do better.

The computational chocolatiers found
that they could achieve a 0.1% savings over
current practice with an equilateral triangle
whose area turns out to be approximately $
1 .99867T. (The exact value for 1 .9986 ... is a |
messy formula involving, for no obvious rea-
son, the square root of 57.) But in fact, all |
that really covers the kugel is a three-leaf s
petal inside the triangle (see figure). That ^
means the tips of the triangle can be cut off, fc
leaving a wrapper of area 1 .8377;i 2 .

The clipped triangular wrapper offers §
another advantage: The length of its perime- “
ter, 5.3503 tc, is shorter than that of any other |
shape the researchers have found. (The 2
square wrapper has a perimeter of 5. 6569 ji; £
the rectangular one, 6k.) So a trefoil wrapper I
would not only save foil, Demaine and col- §
leagues conclude, it would also be cheaper to o

282

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

cut. The potential reduction in the carbon
footprint associated with Mozartkugel mate-
rials and manufacturing, they joke, “partially
solves the global-warming problem and con-
sequently the little- reported but equally
important chocolate-melting problem ”

A Royal Squeeze

In 1850, the great German mathematician
Carl Friedrich Gauss took a shine to a funky
little counting problem: How many ways can
eight queens be placed on a chessboard so
that no two queens attack one another (i.e.,
line up horizontally, vertically, or diago-
nally)? It’s not obvious it can be done at all,
but it turns out there are 92 solutions. Gauss
didn't spot them all, proof in itself that the
problem is a bit of a poser.

Modem computers can easily find all 92,
but mathematicians have upped the ante so
that even Deep Blue would scratch its silicon
head, mainly by making the board larger.
There are, for example, 2,207,893,435,808352
ways of placing 25 nonattacking queens on a
25 x 25 chessboard, a computation completed

Vivat regina. Adding pawns makes a classic chess-
board problem even more queenly.

3 years ago at INRIA.

“There’s a lot of interesting theory behind
these questions,” notes Loren Larson, a chess-
board problem expert in North!' eld, Minnesota.
“They're also nice programming exercises.
They're good examples of backtracking algo-
rithms,” also known as depth- first searches.

In a talk at MathFest, Doug Chatham of
Morehead State University (MSU) in Ken-
tucky described a variant he and collabora-
tors have explored, in which pawns are

NEWS OF THE WEEK

I

allowed on the chessboard. The pawns inter-
rupt the queens’ line of sight, making it pos-
sible for more queens to fit on the board.
How many more queens, they wondered, do
the pawns make possible?

Chatham and crew — MSU colleagues Gerd
Fricke and R. Duane Skaggs, Maureen Doyle
of Northern Kentucky University in Highland
Heights, Matthew WoltY of Pyramid Controls
Inc. in Cincinnati, Ohio, and MSU student Jon
Reitmann — have proved that each additional
pawn permits an extra queen, provided the
board is large enough. For example, with two
pawns, it's possible to get 1 0 queens on a stan-
dard 8x8 board (see figure). In the current
proof, fitting an extra k queens using k pawns
cm an JVx N board requires N to be greater than
25 k, Chatham notes, but adds, '“We believe the
actual minimum sizes are much smaller.”

There are no immediate applications for die
queens-and-pawns problem, Chatham says,
but die original nonattacking-queens problem
has found uses in computer science for parallel
memory schemes and in statistical physics for
particle models with long-range interactions.
“We hope to find similar applications for our
problem,” he says. -barry Cipra

Taking the Edge Off

Math has a lot to say about packing things together. The
abstract problem of cramming, for example, equal-
sized circles into a larger square has applications as
far-flung as error-correcting codes for digital
communications and the physics of granular
materials such as sand. But what if the square
has no edges? A quartet of researchers has
3 shown how packing works in such a borderless space.

^ The space in question is a torus, a shape like the surface

| of an inner tube. To topologists, a torus is equivalent to a
| parallelogram with its opposite edges glued together. On

* the unfurled, flattened-out toms map, anything leaving on
fcj one side immediately reenters from the other, as in many
° video games. William Dickinson of Grand Valley State
j-j University in Allendale, Michigan, and undergraduates
$ Daniel Guillot of Louisiana State University, Baton Rouge, Anna
g Castelaz of the University of North Carolina, Asheville, and Sandi
I Xhumari of Grand Valley have spent the past two summers studying

* circle packings in tori.

5 $ Because a torus has no boundary, the circles are constrained
t only by one another— just as they would be on a patch of regularly
I repeating patterned wallpaper. Dickinson and students classified

the graphs that can result when lines are drawn
connecting centers of tangent circles (red lines
in the figure, below), then set to work analyzing
which ones lead to the densest packings (i.e.,
packings with circles of the largest possible radius).
For five circles — the first truly challenging case —
they found 20 different ways the circles could be
arranged on the torus.

They applied the theory to two particular tori: the
“square” torus formed by connecting opposite
edges of a square, and the “triangular'’ toms, which
starts from a rhombus with a 60-degree angle. Guillot
and Castelaz found the best five-circle packing for
the triangular torus last summer (2007), and
Xhumari did the same for six circles this summer.
Together, the ideas they developed enabled Dickinson
to nail down the densest packing for five circles on
the square toms. It occupies rc/4 or 78.5% of tire square toms, as
compared with 71.1% on the triangular torus (see figure).

“In general, it is very difficult to prove that a particular pack-
ing is optimal,” says Ronald Graham, a circle-packing expert at
the University of California, San Diego. Working without bound-
aries may make proofs easier to come by, he thinks, “but that is
just an impression.” -B.C

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1283

Investigating the Psychopathic Mind

With a mobile brain scanner and permission to work with inmates in
New Mexico state prisons, Kent Kiehl hopes to understand what goes
awry in the brains of psychopathic criminals

ALBUQUERQUE, NEW MEXICO— Kent Kiehl

remembers his first conversation with a
psychopath as if it were yesterday. Kiehl had
just started a graduate program in psychology,
and he intended to study the criminal mind by
interviewing prisoners. His first subject was a
thief who'd made a fortune robbing banks in
North America and lived the high life for
years, renting luxury apartments across
Europe and— if he did say so him-
self — enjoying a great deal of
success with the fairer sex. "Have
you ever had 15 women in one
night?” he asked Kiehl.

The man was behind bars not
because of a heist gone wrong but
because one of his girlfriends was cheating on
him. He tracked her down at a motel room
and burst in with his gun drawn. He shot her
lover, but the man managed to get away. The
woman later testified against him in court. If
he could do it all over again, he told Kiehl, he
would have killed them both. Such stories
fascinate Kiehl, now an associate professor of
psychology and neuroscience at the Univer-
sity of New Mexico and director of Mobile
Imaging Core and Clinical Cognitive Neuro-
science at the Mind Research Network

(MRN) in Albuquerque. "The other 300 or so
psychopaths I’ve interviewed are just as inter-
esting,” he says.

At age 38, Kiehl is embarking on a project
he hopes will unravel the neural basis of psy-
chopathy, a suite of personality' and behavioral
traits that is far more common in violent crimi-
nals than in the general population and is a
strong predictor of repeat offenses. Given the
crime and other societal costs
caused by psychopathic individu-
als, Kiehl says, this group has been
woefully understudied. He intends
to change that. With a custom-built
mobile magnetic resonance imag-
ing (MRI) scanner — roughly $2.3
million of equipment packed into a 1 5-meter-
long trailer — and permission from the New
Mexico governor to work in all 1 2 state prisons,
Kiehl aims to scan 1000 inmates a year.

"We’ll have to see i f he gets that much done,
but if anybody can do it, Kent can,” says Joseph
Newman, a psychologist at the University of
Wisconsin, Madison. "He has big ideas, and he
pursues them energetically.”

Kiehls team conducts hours of interviews
with each subject to assess them for psy-
chopathy, substance abuse, and other mental

health problems. In addition to functional
MRI (fMRI) experiments to investigate neu-
ral activity during various tasks, they’re also
collecting anatomical images of the brain and
DNA samples that could eventually be used
to search for genetic risk factors — all with die
prisoners’ full consent and cooperation and
all to be used solely for research. Kiehl’s
research is funded by four R01 grants from
the National Institutes of Health, which pay
about $900,000 a year in direct costs; MRN
paid for the scanner.

Depending on what he finds, Kiehl’s work
could raise a host of legal and ethical ques-
tions. Could brain scans or blood tests one
day improve on the personality profiles and
other low-tech methods now used to assess
the degree of risk a prisoner poses to society?

If so, how should they be used? Could a better
understanding of the psychopathic brain alter
the way we think about the culpability of cer-
tain criminals? Could it point the way to inter-
ventions that prevent recidivism?

We’ll never know unless we do the
research, Kiehl says: "We just have no idea
how their brains are different, how they got
that way, and how we might be able to treat |
the condition.”

S

Local boy does bad

Kiehl s interest in psychopathy goes back to §
his childhood. He grew up in a middle-class |
neighborhood in Tacoma, Washington, not u

Online

sciencemag.org

Kl Podcast interview
*2.? with the author of
this article.

1284

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

NEWSFOCUS

far from the boyhood home of serial killer Ted
Bundy. While Kiehl was in grade school,
Bundy was on a nationwide rampage, killing
dozens of young women. Kiehl’s father was a
newspaper editor at the time, and Bundy’s
exploits were a common topic of discussion
at the family dinner table.

Bundy exhibited several defining traits of
psychopathy. He was cunning and manipula-
tive, often donning disguises or feigning
injury to lure women into a vehicle, and his
preferred method of killing — crowbar blows
to the head — as well as his proclivity for sex
with his dead victims suggest a stunning lack
of empathy. “Why would someone from my
neighborhood end up being so bad?” Kiehl
remembers wondering at the time.

By the time Bundy was executed in
Florida in 1989, Kiehl was fantasizing about
becoming a professional athlete. He entered
the University of California (UC), Davis, that
year after being recruited to play on the foot-
ball team. Solidly built at 6’2”, Kiehl still
exudes an athlete’s self-confidence. On a
recent afternoon, he collected on a S100 bet
with his lab manager over how far he could hit
a golf ball. “I bet I could hit a ball farther than
Tiger Woods,” he boasted.

When a knee injury forced Kiehl to
reconsider his life goals, he recalled his fas-
cination with Bundy and began getting more
interested in neuroscience. He rotated
through the laboratories of several UC Davis
neuroscientists, setting his sights on gradu-
ate work with psychologist Robert Hare at
the University of British Columbia (UBC) in
Vancouver, Canada. Hare is a preeminent
psychopathy researcher who in 1 980 pub-
lished the first version of what has become
the main tool for diagnosing
psychopathy. In its current
incarnation, the Psychopathy
Checklist-Revised (PCL-R)
scores subjects on 20 traits indi-
cative of psychopathy, including
S callousness, impulsivity, and a
| history of behavioral problems.

| People in the general population
5 typically score a four or five on
| the 40-point scale, Hare says. A
| score of 30 is widely used as a benchmark
g for psychopathy.

a Psychopathy is not listed in the American

1 Psychiatric Association’s Diagnostic and Sta-

2 tistical Manual of Mental Disorders, 4th ed.
| ( DSM-IV ). The DSM-IV diagnosis of anti-
g social personality disorder captures some of
£ the external manifestations of psychopathy,
t including impulsivity and antisocial behavior.
§ but ignores personality traits such as glibness,
u callousness, and lack of remorse that are

scored by the PCL-R. Studies with prison
populations have found that roughly 20%
(slightly more or less, depending on the secu-
rity level of the prison) of inmates qualify as
psychopaths. Incarcerated psychopaths have
committed an average of four violent crimes
by the age of 40, Kiehl says. More than 80%
of those who are released from prison commit
another crime, usually a violent one, within
3 years, compared with 50% for the overall
prison populatioa “Psychopathy is the single
best predictor of violent recidivism,” says
Kiehl, who hoped to collaborate with Hare to
study the brains of psychopathic criminals.

But Hare wasn’t interested in taking him
on. “I had a lot of really outstanding students
applying to work in my lab, and his grades
weren’t particularly great,” Hare says. Not
one to give up easily, Kiehl launched a cam-
paign that included a barrage of recommen-
dation letters from UC Davis faculty mem-
bers; he also drove through a snowstorm
from Tacoma to Vancouver to hand-deliver a
few bottles of California wine that he knew
Hare would appreciate. “That did it,” says
Hare. “He wore me down.”

An emotional problem?

Long before fMRI scanners came along,
researchers suspected that psychopathy
springs from a defect in emotional process-
ing in the brain. Several of the disorder’s
signature traits hint at this, as do early
studies that found blunted physiological
responses — by measures such as heart rate
and skin conductance — to emotionally
evocative photographs in psychopaths.

Such abnormalities cast obvious suspi-

cion on the amygdala, the hub of emotion in
the brain. In the first fMRI study of psy-
chopathy, published in 200 1 in Biological
Psychiatry, Kiehl and UBC colleagues found
reduced amygdala activity in psychopathic
criminals compared with nonpsychopathic
criminals in response to emotionally charged
words. A malfunctioning amygdala is likely
to be one crucial factor in psychopathy, says
James Blair, a cognitive neuroscientist at the
National Institute of Mental Health in
Bethesda, Maryland. Human and animal
studies have shown that the amygdala is
essential for learning to avoid behaviors with
unwanted outcomes, he notes. By preventing
children from learning to avoid actions
that harm other people, faulty wiring in
the amygdala could derail normal social
development and contribute to the callous,
unemotional traits seen in psychopaths, he
proposes. In the June issue of The American
Journal of Psychiatry, his research group
reports that children with callous, unemo-
tional traits have less amygdala activity than
other children when viewing photos of fear-
ful facial expressions.

Other researchers question whether the
amygdala is really the source of the problem,
however. Newman, for example, has long
argued on the basis of behavioral evidence
that deficits in regulating attention may be
the central issue for psychopaths. “Once
they start paying attention to some goal they
want, they ignore cues that would otherwise
activate the amygdala,” he says.

Kiehl takes an even broader view. He sus-
pects that psychopathy involves disruptions
to a network of “paralimbic” regions in the
brain’s temporal and frontal lobes that
contribute to emotion, attention, decision-

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1285

| NEWS FOCUS

making, and other cognitive functions.
Resolving some of the confusion about
which cognitive processes — and which brain
regions — are dysfunctional in psychopathy
is a major goal of his neuroimaging work in
New Mexico.

But neuroimaging has limitations
(Science, 13 June, p. 14121. The behaviors
that can be studied inside an fMRI scanner,
for example, are necessarily simplified and
artificial. Proving that any given neural
abnormality that shows up in imaging actu-
ally contributes to psychopathic traits and
behavior in real life is never easy, says
Adrian Raine, a clinical neuroscientist at the
University of Pennsylvania. And then there’s
the chicken-and-egg problem. “Is it leading
a violent, psychopathic way of life that
causes the structural and functional impair-
ments we find, or is it the other way
around?” Raine asks. “It’s going to be hard
to answer that very important questioa”

Prison-bound

On a blazing hot day in late July, Kiehl’s
mobile scanner was parked inside the gates
topped with razor wire at the Youth Diagnos-
tic and Development Center in Albu-
querque. From the outside, the mobile
resembles any trailer you’d see on an 18-
wheeler, albeit cleaner than most. Kjehl
spent a year working with engineers at
Siemens to design it and ensure that the
scanner's magnetic field would remain sta-
ble in different locations. Inside, the mobile
looks like an ultra-high-tech recreational
vehicle. The scanner sits at one end, its mag-
netic cylinder a pale blue doughnut extend-
ing from floor to ceiling. Flat-screen moni-
tors adorn the walls in the adjacent control
room, and next to that a small sitting room
contains a stack of magazines for the benefit
of a corrections officer who waits here while
a juvenile prisoner gets scanned.

All experiments are off-limits to the
media, in part because of concerns about the
privacy of prisoners but largely because of a
bad experience Kiehl had in Canada. A televi-
sion network broadcast an interview with one
of his research subjects that was edited to
make the guy seem even scarier than he was,
Kiehl says. When the inmate was denied
parole a short time later, he threatened to kill
any other inmates who participated in Kiehl Is
research; he also threatened to hit Kiehl with a
chair. Now Kiehl says he won’t jeopardize his
staff by allowing the media to watch experi-
ments or interview inmates.

Despite the nature of some of their sub-
jects’ crimes, Kiehl’s students and postdocs
say that they've never felt threatened. “They

tend to really like us,” says postdoc Matthew
Shane. “They enjoy any excuse to talk with
someone from outside the prison.”

In one of the first studies using die mobile
scanner, Kiehls postdoc Carla Harenski and
colleagues investigated how the brains of
adult male prisoners respond to morally
charged photographs, such as an image of a
man holding a knife to a woman’s throat. The
inmates also rated the severity of the “moral
violation” depicted in the photographs on a
five-point scale. Those who gave high scores,
suggesting greater sensitivity to moral viola-
tions, tended to have more activity in the
superior temporal sulcus, a region implicated
in previous studies of moral judgments, the
researchers reported at an April meeting of

Neural roots. Kiehl suspects that disruptions to
paralimbic brain regions (light areas) underlie
psychopathy.

the Cognitive Neuroscience Society. The
team has subsequently scanned a bigger sam-
ple of prisoners and is investigating whether
activity in this and other brain regions differs
between those who are psychopathic and
those who aren’t.

Into the courtroom?

Such differences in brain activity within
prison populations could potentially prove
useful in assessing the risk posed by individ-
ual criminals, perhaps as a supplement to the
PCL-R, Kiehl says. That checklist is cur-
rently used in dozens of countries. Depend-
ing on the jurisdiction, PCL-R scores are
considered during sentencing and parole
hearings. Some prisons use them, along with

other factors, to determine security measures
and treatment options.

Whether brain scans will ever prove useful
in such settings depends on whether they add
any predictive power, says Walter Sinnott-
Armstrong, a philosopher at Dartmouth
College and co-director of the MacArthur
Foundation's Law and Neuroscience Project
in Hanover, New Hampshire. Not everyone is
optimistic. “It’s not some sort of crystal ball
that’s going to tell you who's going to reoffend
in 5 years’ time,” says Essi Viding, a cognitive
neuroscientist at University College London.
She also questions the practicality of the
approach, given that MRI scans cost $ 1000 or
more apiece and require substantial technical
expertise. Even so, research on the neural
basis of psychopathy could have important
legal implications, says Sinnott-Armstrong.
For example, he says, if future research points
to a diminished moral capacity due to a neuro-
developmental defect, that could be relevant
in court, where a defendant's understanding of
the wrongfulness of his actions has a bearing
on the verdict.

Kiehl gets impatient with such hypotheti-
cals. For him, the ultimate question is how
best to intervene — ideally, early in life before
psychopathic traits become ingrained. The
conventional wisdom is that psychopathy is
unbeatable, but that’s based “more on clini-
cal lore than solid research,” says Michael
Caldwell, a psychologist at the Mendota
Juvenile Treatment Center and the Univer-
sity of Wisconsin, Madison. One widely
cited study found that psychopaths who par-
ticipated in a treatment program in the 1970s
actually did worse than those who didn’t,
Caldwell says. But given that the treatment
regimen involved nude encoimter groups and
LSD, those findings should perhaps be taken
with several grains of salt. Kiehl says he’s
been buoyed by a recent series of papers by
Caldwell and colleagues that suggest that tar-
geted interventions, including cognitive
behavioral therapy and family counseling,
with juvenile offenders with psychopathic
traits can prevent future crimes.

Caldwell, Newman, and other veteran
psychopathy researchers say that they’re
encouraged to see Kiehl’s project getting off
the ground because public support and fund-
ing for psychopathy research has been hard
to come by in the past. “If someone is cruel
and always out for himself, it’s not something |
that engenders sympathy, concern, and the S
desire to understand it,” says Newman. “My S
view is that it’s a really important disorder |
that needs to be understood.” Kiehl says he §
couldn’t agree more.

-GREG MILLER 3

1286

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

particles at energies
seven times the previ
ous record.

The Overture Begins

Next week, physicists at the European particle
physics lab, CERN, will fire up the world's biggest
atom smasher. Expectations are skyhigh, but
discoveries may still be years away

Fourteen years ago, scientists at the European
particle physics laboratory, CERN, near
Geneva, Switzerland, had only plans for a new
highest energy particle smasher. Now, thanks
to the efforts of thousands of people, they
have a gargantuan machine, the S5.5 billion
Large Hadron Collider (LHC), which
stretches through a 27-kilometer ring of tun-
nel between Lake Geneva to the east and
France’s Jura Mountains to the west ( Science ,
23 March 2007, p. 1652). “It seemed like an
enormous mountain to climb, that’s for sure,
back when we didn’t have even a single mag-
net,” says CERN’s Lyndon Evans, w'ho has led
the project since its inception.

Evans says he’s had moments of despair.
In 2004, a manufacturing error forced work-
ers to rip out and rebuild 3 kilometers of the
high-tech plumbing that carries frigid liquid
helium to the accelerators superconducting
magnets. In 2002, cost overruns led officials
to delay the completion of the LHC by a year.
But now, as researchers test the LHC’s myr-
iad subsystems, “it really feels like an old
friend,” Evans says. “It acts exactly like it is
supposed to act.” Physicists around the world
hope their ami, the most complex scientific
apparatus ever built, continues to behave
next week when, for the first time — provided
that lawsuits do not force a delay (see side-
bar, p. 1291) — researchers try to circulate
| particles through its twin rings.

«• In the quest to unravel the universes inner
I workings, the 10 September start-up of the
u LHC marks the beginning of a new age of

exploration. The collider should bag the long-
sought Higgs boson, the missing link in physi-
cists’ “standard model” of the known particles
and the one thought to give the others their
mass. It could glimpse a slew of new particles,
such as those predicted by a scheme called
supersymmetry, or even reveal new dimen-
sions of space. Other colliders hammered out
how the standard model is structured; the LHC
should answer deeper questions about why the
model is as it is, says Gordon Kane, a theorist
at the University of Michigan, Ann Arbor.
“The LHC is a ‘why’ machine,” he says.

But answers most likely won’t come right
away, cautions CERN’s Peter Jenni,
snnkesnerson for the

What a blast! In this simulation, a Higgs boson is
born and decays inside the ATIAS particle detector.

2500-member team working with the 25-
meter-tall, 45-meter-long ATLAS particle
detector— one of four big detectors the LHC
will feed. “People should definitely not take it
for granted that big tilings will happen imme-
diately,” he says. If all goes well, the LHC will
start smashing particles in October, and oddi-
ties could jump out right away. More likely, it
will take a few years for the LHC to clinch the
discovery of the Higgs or something even
stranger. Still, after 3 decades in which the
standard model has answered every’ question
asked at particle accelerators, physicists are
eager to see something really new.

First off, look for something old

Like all colliders, the LHC aims to produce
fleeting bits of matter not seen in the every-
day world. As Einstein noted, energy and
mass are equivalent. So physicists can gener-
ate heavier exotic particles by smashing
known ones together with sufficient energy.
Blasting protons into protons at energies
seven times as high as the previous record,
the LHC could cough up new particles more
than 1000 times as massive as a proton. But
first, researchers will simply search for
familiar standard-model particles.

Ordinary matter consists of particles
called “up quarks” and "down quarks,” which
combine to make the protons and neutrons in
atomic nuclei; the electrons that make up the
rest of the atom; and wispy particles called
neutrinos that emerge in a particular type of
nuclear decay. This first family of particles is

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1287

NEWSFOCUS

Researchers, Place Your Bets!

The days before the start-up of the Large Hadron Collider (LHC) should be

filled with quiet contemplation and reverence for the adventure to come, says

physicist Maria Spiropulu. "Now is not the time to speculate," says the experimenter at

the European particle physics lab, CERN, near Geneva, Switzerland. "We should be silent and

respectful and wait for the data to come."

Or not. Many physicists seem to think that now is precisely the time to guess at what CERN's
great particle smasher might find. And some are even willing to put their money where their

favorite theoretical models are and
wager on their expectations.

Tommaso Dorigo, an experimenter at
the University of Padua in Italy, doubts
that the LHC will find evidence of super-
symmetry, a theoretical scheme that pre-
dicts a massive "superpartner" for every
known particle in physicists' current
"standard model." In the past 10 or 15
years, extremely precise measurements of
standard-model particles have indirectly
undermined the viability of the notion,
on't believe in the thing," he says. Dorigo has
r physicists that, after the LHC has accumu-
data, it will see no sign of supersymmetry,
o has bet that the LHC will see no clear devi-
model of any kind, explains Jacques Distler,
sity of Texas, Austin, who has $750 of the
) professional gambler, Distler says he took
the bet because it is so open-ended that he likely can't lose. "History has always been, you explore
a new energy range and you see something new," he says.

For some, not having a bet bespeaks the strength of their predictions. Gordon Kane, a theorist at the
University of Michigan, Ann Arbor, says he would gladly wager that the LHC will find supersymmetry, but
"nobody I know will bet against it." Stuart Raby, a theorist at Ohio State University in Columbus, also says
he can't find anyone who will take such a bet. To which Distler says, "I wonder how hard they tried."

The general public can get into the game, too. Online gambling sites and prediction
exchanges such as lntrade.com, Hubdub.com, and NewsFutures.com are taking bets on when the
Higgs boson will be discovered, whether theTevatron collider at Fermi National Accelerator
Laboratory in the United States will see it first, and related questions. -A.C.

High rollers. Tommaso
Dorigo (below) wagers
that the LHC will see
nothing new. Jacques
Distler disagrees and
expects to pocket $750 of
Dorigo's money.

Dorigo says. "I realized I d
bet $1000 with two othe
tated a certain amount of
More precisely, Dorig
ations from the standard
a theorist at the Univer:
action. Like a calculating

flanked by two sets of heavier, unstable rela-
tives. That means there are also strange and
charm, top and bottom quarks; the electron
has beefier cousins called the muon and the
tau lepton; and there are two more “flavors” of
neutrinos. Still other particles convey forces:
Photons carry the electromagnetic force, the
massive W and Z bosons convey the weak
nuclear force, and gluons make up the strong
nuclear force that binds protons and neutrons.

Tracking such familiar particles will enable
experimenters to calibrate their immensely
complex devices, saysTejinder “Jim” Virdee, a
physicist at Imperial College London and
spokesperson for the 2900-member team
working with the 12,500-ton CMS particle
detector, ATLAS’s rival. (The LHC’s two other
detectors, ALICE and LHCb, won’t search
directly for new particles but will do more
specialized work.) For example, a Z boson can

decay into a muon and an antimuon, so by
studying Z’s physicists can measure how well
they spot those particles.

Such studies also set the baselines from
which to search for something new, Virdee
says. “If you see something [unusual], the first
question everyone is going to ask is, ‘Do you
also see the other things you expect?’ ” he says.
“You have to be able to say ‘yes’ before you can
claim anything new.”The LHC should produce
a smidgen of data between October and
December, when it will shut down for the win-
ter, and experimenters will use it primarily to
“rediscover" the standard model.

The Higgs: Wait a couple of years

Of course, experimenters will also keep an
eye out for new things, such as the elusive
Higgs bosom That oddball particle solves a
serious problem with the standard model: The

theory goes mathematically haywire unless
particles have no mass. The “Higgs mecha-
nism” sidesteps that problem by generating
mass through the interactions of the otherwise
massless particles themselves. It assumes that
empty space is filled with a field a bit like an
electric field that drags on particles to give
them inertia, the essence of mass. Just as an
electric f eld is made up of photons, the Higgs
field consists of particles — Higgs bosons —
that can be ripped out of the vacuum.

But finding the Higgs may not be easy. It all
depends on how much the particle weighs, says
Karl Jakobs, a physicist at the University of
Freiburg in Germany and physics coordinator
for ATLAS. The standard model does not pre-
dict how heavy the Higgs should be. If it weighs
between about 200 and 500 times as much as a
proton, then it should stick out fairly clearly. In
that case, experimenters might collect enough
data to find it by die end of 2009. Jakobs says,
although analyzing the data could take months
longer. But previous searches and indirect
inferences suggest that the Higgs is lighter —
definitely more than 121 times as massive as a
proton but probably less than 1 70 as massive as
that benchmark. If the Higgs is that light, then it
could take until 201 2 or later to f nd it.

The difference is that if the Higgs is heavy
enough, it should decay in a distinctive
way — into two hefty Z’s that both decay into
a muon and an antimuon. But if the Higgs is
too light for that, then researchers will have
to look for it decaying into combinations
such as a pair of photons. So many photons
will be produced in a typical LHC collision
that sifting out the Higgs’s signal from the
clutter will take lots of data.

Most physicists say that they are sure to
find the Higgs or something even weirder,
because without it the standard model again §
breaks down mathematically at the energies I
the LHC will reach. Ironically, finding only the |
Higgs boson would disappoint many, as it ^
would leave physicists nothing to puzzle over. S
“The worst scenario for me is that you start j3
running and you see no evidence of deviations g
from the standard model, and after 2 or 3 years o
you see evidence of a standard-model Higgs §
and nothing else,” Jakobs says. The Higgs o
would be the last brick in the standard model . It |
alone would leave physicists facing a concep- 2
tual wall and could signal the end of die field. ^

Spotting signs and nailing discoveries §

Most physicists expect to f nd much more at £
the eneigies the LHC will explore. New forces a
might emerge, or quarks themselves could turn £
out to consist of other particles. More specula- 5
tively, space may have additional curled-up §
dimensions that might be pried open, or the o

1288

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag. 0 r 9

NEWSFOCUS

LHC might make tiny black holes, which
would tie together the realms of quantum
mechanics and gravitational physics.

Perhaps the most favored idea is super-
symmetry, a scheme that posits a heavier, un-
observed “superpartner” for every particle in
the standard model. Seemingly profligate in
its complexity, supersymmetry would help
solve a number of fundamental, albeit eso-
teric, problems in the standard model. For
example, it helps unify the three forces in the
theory, a prerequisite to formulating a theory
in which all forces, including gravity, are dif-
ferent manifestations of a single master force.
Supersymmetry might also provide the myste-
rious dark matter whose gravity holds the
galaxies together, as the least massive super-
partner would be a heavy particle that would
interact with ordinary matter essentially only
through its gravity.

Supersymmetry might be very easy to see
at the LHC, some say. “We predict a signature
that they could see with five events,” says
Michigan’s Kane. “They could see it in the
first week of running in October.” Generally,
collisions producing the undetectable least
massive supersymmetric particle would look
lopsided, with a spray of ordinary particles
shooting out one side of the particle detector
and the supersymmetric particle zipping out
the other side without leaving a trace.

But although spotting those events may
be easy, proving that they’re evidence of
supersymmetry and not something else may
be tough, says CERN’s Paraskevas Sphicas,
physics coordinator for CMS: “The catch is
that the signature is so complex that we
would have to do a lot of analysis to under-
stand it.” In fact, clinching the case for super-
symmetry could take several years.

First, however, physicists must get the
machine up and running. Researchers have
already succeeded in injecting protons into
each of its countercirculating rings. On 10
September, they’ll try to coax the beams all
the way around the rings at a very low eneigy.
They’ll then aim to increase the beam energy
to 70% of the ultimate goal and the beam
intensity to 1/1000 the design standard before
beginning collisions in several weeks’ time.
Next year, the LHC should smash a billion
particles each second at full energy.

The first collisions will mark the beginning
of the real fun for experimenters. Some say that,
although they have some pretty good ideas, they
don’t really know what to expect. “I want sur-
prises,” says CERN’s Maria Spiropulu, an
experimenter working on CMS. She may well
get them, although she and her colleagues may
have to wait just a bit longer. -ADRIAN CHO

Bracing for a Maelstrom of Data,
CERN Puts Its Faith in the Grid

Researchers have hammered out new networking tools to store the LHC's instrument
readings and make them available to physicists worldwide

After the Large Hadron Collider (LHC) pow-
ers up next week, the physicists and engi-
neers who built the machine and its detectors
won’t be the only ones nervously waiting for
its two beams to collide for the first time. Just
as anxious will be the researchers charged
with taking the flood of data that the LHC

will produce and processing it, storing it, and
making it available for physicists to study the
world over. The LHC is expected to produce
15 petabytes (15 million gigabytes) of data
every year. Dedicated fiber-optic lines have
been laid down to whisk the data away from
CERN to some 250 other physics labs in 50

countries worldwide, where about 100,000
PC processors are ready and waiting to
receive them.

At the beginning of this decade, CERN’s
information technology (IT) department
decided to handle the LHC’s torrent of data
using a novel computer architecture known
as a grid. A grid is a way of using the Internet,
just as the World Wide Web and e-mail are.
But the technology has not developed as fast
as particle physicists had expected. CERN
researchers believe they have ironed the
wrinkles out of their system, dubbed the
LHC Computing Grid (LCG), but nagging
doubts remain.

“By an order of magnitude, this is the
biggest grid [yet assembled],” says John
Gordon, deputy director of GridPP, the
United Kingdom’s contribution to the LCG.
“I’m reasonably confident that the grid is
ready for data.” But Les Robertson, head of
the LCG project from its inception in 2001
until the beginning of this year, adds a note of
caution: “It’s very difficult. There’s no real
data, and real users are not active. A live test
will only come when [real] data starts to
flow.” He adds: “This is what we will use.
There’s no fallback.”

Fifteen petabytes is an enormous amount
of data. To store it all on CDs would require a
stack of disks 20 kilometers high — more than
four times the height of Mont Blanc, Europe’s
tallest mountain. When CERN’s IT experts
began planning how to handle data from the
LHC in the late 1 990s, they soon realized that
it would not be feasible to do it all at CERN. It
wasn't clear that Geneva’s electricity supply
could power enough computers to do the job,

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1289

NEWSFOCUS

processing power, storage,
scientific instruments,
simulation, and so on.

It is called a grid in a
deliberate analogy to
the electricity grid.
When you plug a toaster
into a socket, you don’t
know how the electricity
was produced or how it
traveled to yoa Similarly,
with a computing grid, a
researcher can use a stan-
dard PC interface to request a
job to be done. Then a program
known as middleware takes over,
marshalling resources from multiple
sites across the Internet. Raw instrument
readings may be taken from a database in
Europe and processed by a supercomputer in
the United States; the manipulated data may
be stored in China and then put through a visu-
alization program in Japan before being
returned to the researcher — who sees only the
results, not the journey that got them there.

A major difficulty in setting up a grid
arises from the “firewalls” that institutions
erect to protect their computers from un-
authorized access. It’s a challenge getting the
differing architectures and security arrange-
ments of all the institutions in a grid to work
together and trust one another. Each job trav-
els around with “certificates” confirming
that the user who requested the job has the
authority to use the resources. To make grids
work, there are “many hurdles, social, politi-
cal, and technical,” says particle physicist
David Britton of the University of Glasgow
in the UK.

As particle physicists learned more about
grids, the CERN team decided that the
approach offered a more flexible way to han-
dle the LHC’s mountain of data, says Robert-
son. Although the same basic layout of tier- 1
and tier-2 centers remains, it is no longer a
rigid structure like the spokes of a wheel, with
users tied to their local tier-1 center. The pro-
duction process of disseminating the LHC

data is handled by the grid, and
researchers can get hold of the
data they want without knowing
where they are or what passwords
they need to get access to them.

Since that decision was made
in 2001, dedicated high-speed
fiber-optic lines have been built
between CERN (tier-0) and the
tier-1 centers. Beyond that, the
normal Internet provides the
infrastructure. Particle physicists
in each participating country
have built up the LCG with funding from
their respective governments for computer
resources to add to the grid. Within the Euro-
pean Union, national grid efforts for research
have been linked to form the Enabling Grids
for E-Science (EGEE), which forms the
backbone of the LCG in Europe. That role is
performed in the United States by the
National Science Foundations Open Science
Grid. Other smaller grids, such as GridPP,
Scandinavia’s NorduGrid, and Italy’s INFN
Grid, have also been woven into the LCG.

In February and May this year,
researchers carried out two major trials of the
system, sending simulated data from the
LHC detectors themselves through CERN’s
tier-0 hub out to tier- Is and tier-2s. Britton
says the February test was “quite successful,
. . . better than we hoped,” although they man-
aged only a couple of days running data from
all four detectors simultaneously. Much fine-
tuning was done before the May dry run, and
as a result they ran the four detectors together
for the entire month. “We tested the whole
chain, and most things stood up,” says Gor-
don. Some bits of software didn’t behave as
expected, he says. In addition, the tier- 1 cen-
ter at Amsterdam had trouble keeping its
computers cool, while the U.K. tier-1 at
Rutherford Appleton Laboratory near
Oxford suffered a small power failure. "But
it was successful because we caught up,"
Gordon says.

Researchers say that almost all the com-
puting resources needed for full LHC opera-
tion are now in place, and they are confident
that the production side of the operation —
transmitting, processing, and storing LHC
data as it’s produced — will go as planned.
The thing that still gives them the odd
sleepless night is what will happen when the
LHC starts producing some interesting
physics. Suddenly, thousands of physicists
across the globe who have patiently waited
years for these data will log onto the grid and
request jobs. Grid experts refer to such use as
“chaotic” because of its unpredictability.
“It’s definitely an unknown still ” says

Trickle down. Beginning as a way for
hundreds of physics labs to divide the
work of processing and archiving LHC
data, the global "grid" evolved into a
universal tool kit for particle physicists
to share and study results.

and in any event, CERN couldn’t afford them:
All of the LHC budget was being spent on the
machine itself. “It was easier to get resources
that were already available at computer cen-
ters,” says Robertson.

At f rst, the CERN team set about design-
ing an architecture in which, as the LHC
detectors chum out data, the information
would be archived in its raw state at CERN
while simultaneously being streamed out to
10 or so large physics labs elsewhere in the
world. At these tier-1 centers, some process-
ing of the data would be done; then it would
be archived again and some data would be
farmed out from each tier- 1 center to 1 0 or 20
tier-2 centers. In this way, the work of pro-
cessing and archiving data is shared among
particle physics labs around the world. The
scheme would have worked, but it lacked
flexibility, and the researchers soon heard
about something better.

In the mid-1990s, Ian Foster and Steven
Tuecke of Argonne National Laboratory in
Illinois and Carl Kesselman of the California
Institute of Technology in Pasadena had
devised the idea of a grid. Whereas the Web is
essentially a system for moving data around
with limited processing for tasks such as
searching, a grid aims to share everything:

1290

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

NEWSFOCUS

Is the LHC a Doomsday Machine?

Even for a car ad, the pitch on the radio was hard to ignore: an "end-of-the-
world sale," offering a 30% discount and $1000 cash back on new auto-
mobiles. "Buy yourself something really frivolous," it urged. The reason:
Miniature black holes created by the Large Hadron Collider (LHC) might soon
touch off an unstoppable chain reaction that would blow up Earth.

Brad Benson, the New Jersey Hyundai dealer behind the ad, isn't really
worried about the fate of the planet. "I’m a National Geographic kind of a
guy," he told Science. "I love reading about this kind of stuff." But as the
$5.5 billion particle smasher prepares to carry its first beam next week, some
people seethe machine as a threat. A handful of physicists and others have
proposed an array of dangerous entities that could be created in the minus-
cule fireball of a particle collision — including microscopic black holes,
strange matter that is more stable than normal matter, magnetic monopoles,
a different quantum-mechanical vacuum, and even thermonuclear fusion
triggered by a stray beam. Discussion forums on the World Wide Web sizzle
with rants against arrogant scientists who meddle with nature and put us all
at risk. And a few groups have sued to stop the LHC.

In March, Walter Wagner, a nuclear physicist based in Hawaii, and Luis
Sancho filed for a restraining order and injunction against the LHC in the
U.S. District Court of Hawaii. This week, Wagner
is due to appear to fight a motion from the U.S.

Department of Energy to dismiss the case.

Meanwhile, late in August a European
group filed a complaint with the European
Court of Human Rights (ECHR) for an emer-
gency injunction to halt the switch-on. On 29
August, after 3 days of deliberation, the court
declined to grant the injunction. An ECHR
spokesperson says the plaintiffs can continue
to pursue the complaint, but given the number
of cases on the court's files it may take as long
as 3 years to decide on its admissibility alone.

"The only serious solution is not even to start
the [LHC] project," says Markus Goritschnig,
spokesperson for the ECHR complaint. "We will
continue the case," he adds.

The LHC is not the first particle collider to face campaigns over its safety.
In 1999, Wagner sued to stop the Relativistic Heavy Ion Collider (RHIC) at
Brookhaven National Laboratory in Upton, New York. The case was dis-
missed in 2000, and RHIC began operating the same year. To forestall sim-
ilar campaigns against the LHC, which part of the time will collide heavy
ions at even higher energies, CERN commissioned five independent physi-
cists and one CERN staffer to assess the dangers of the new machine. Their
conclusion, published in 2003: "We find no basis for any conceivable risk."
A second panel, the LHC Safety Assessment Group (LSAG), came to the same
conclusion in a report published this June.

The doom mongers do have one thing right: The LHC just might create
black holes. According to Einstein's theory of general relativity, energy
warps space and time. So by smashing protons together with unprece-
dented vigor, the LHC might cram enough energy into a small enough vol-
ume to create pinholes in the universe — miniature black holes. If space
has three dimensions, even the energies reached by the LHC will be about
a million billion times too low. However, string theory — which assumes
that every fundamental particle is in fact an infinitesimal vibrating
string — predicts that space has more dimensions curled into tiny loops. If
some of them are curled loosely enough, then the energy threshold may
tumble to within the LHC's reach, some theorists have argued.

But such tiny black holes should quickly evaporate into ordinary parti-
cles. At the least, they must be able to decay back into the particles that
created them. They should also decay through "Hawking radiation," which
comes about when, thanks to quantum uncertainty, a particle-antiparticle
pair pops out of the vacuum and one partner falls in the hole while the
other shoots outward.

LHC opponents point out that no one has ever observed Hawking radi-
ation, and they fear that the black holes will grow and gobble up more and
more matter. German physicist Rainer Plaga, in a paper cited in the ECHR
complaint, theorizes that black holes could both grow and radiate
intensely, doing as much damage through
radiation as they do by eating everything in
sight. In another cited paper, Otto Rossler, a
theoretical chemist at the U niversity of Tubin-
gen in Germany, begins with an unusual —
and, physicists say, wrong — interpretation of
general relativity to argue that minuscule
black holes should be stable and may form
tiny radiation-spewing quasars.

All those scenarios are based on dodgy
reasoning, says Jonathan Ellis, a theorist at
CERN. Besides, he says, Earth, the sun, and
other celestial bodies are constantly bom-
barded by cosmic particles with energies far
higher than the LHC will reach. As the LSAG
noted in its report: "This means that Nature
has conducted the equivalent of about a
hundred thousand LHC experimental programmes on the Earth
already — and the planet still exists."

Physicists may have unwittingly helped foment panic by talking too
glibly about black holes, Ellis notes. "Maybe we should be more careful
with our rhetoric," he says. "For example, we talk about recreating the big
bang, and people think, 'Oh my God, they're going to recreate the big
bang!' " Of course, physicists don't aim to literally return the universe to
its fiery birth, just to mimic those conditions in fleeting particle collisions.
Alas, that less sexy line isn't going to catch anyone's attention, as any
good car salesman can tell you. -D.C. AND A.C.

3

i

s' Gordon. Britton agrees. “It will be a chal-
§ lenge to the grid because there will be a large
> number ofless expert users” he says. “We’ll
I have to leam how to help users in this type of
| environment.”

| LCG researchers were surprised that it
£ has been this hard to develop the grid. At the
| outset, they expected it to evolve as the World

1 Wide Web did: After CERN invented it,

2 industry took the ball and now provides the

Web as a service to researchers and the pub-
lic alike. Although some companies, includ-
ing Amazon, are starting to provide gridlike
services commercially — the buzz phrase is
“cloud computing” — the LCG researchers
had to develop much of the new system as
they went along. “We hoped the grid would
be a service by now, but it hasn’t happened,”
says Tony Cass, head of fabric infrastructure
and operations in CERN’s IT department.

Britton acknowledges that it was a risk
going down the grid route, but he says the
particle physics community looked at the
technology, assumed it would develop, and
assumed they could make it work in the time
available — just as they did with the rest of
the LHC. “That’s exactly what particle
physicists have to do: push things beyond
the current envelope.”

-DANIEL CLERY

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1291

LETTERS I BOOKS I POLICY FORUM I EDUCATION FORUM I PERSPECTIVES

LETTERS

edited by Jennifer Sills

i

5

i

.

i

t

1

I

Reading Between the Number Lines

IN THEIR REPORT "LOG OR LINEAR? DISTINCT INTUITIONS OF THE
number scale in Western and Amazonian indigene cultures” (30 May,
p. 1 2 1 7), S. Dehaene et al. investigate how Mundurucu Indians of the
Amazon map numerosity judgments on a line segment. They conclude
that the concept of a linear number line is a product of culture and for*
mal education and that “the mapping of numbers onto space is a uni-
versal intuition and that this ini-
tial intuition of number is loga-
rithmic.” While I fully agree with
the former statement — which I
have defended elsewhere (7) — I
disagree with the latter.

First, if the intuition of map-
ping numbers onto space is as
fundamental as the authors claim,
we should expect ubiquitous

Ancient math. By providing insight
into the math used in ancient cultures,
artifacts such as this 4000-year-old clay
tablet can help to distinguish learned
mathematical concepts from those that
are intuitive.

manifestations of number lines — linear or logarithmic — in early arith-
metic in Mesopotamia. Egypt, China, and Mesoamerica. But no such
evidence exists. Indeed, 4000-year-old clay tablets show that
Babylonians developed a sophisticated knowledge of arithmetical
bases, fractions, and operations apparently without the slightest refer-
ence to number lines (see photograph) (2). The number line is a com-
plex idea that appears to have been introduced as late as 1685 by John
Wallis (J). Second, if humans’ initial intuition of“number” is logarith-
mic, we wouldn’t have had to wait until the 1 7th century to see the
invention of logarithms through Napier’s painstaking
work. These late inventions are inconsistent with the
authors’ claim that “mathematical” objects may find their
ultimate origin in basic intuitions internalized through
millions of years of evolution. The Report seems to be
less about mathematical concepts and more about the role
a line segment can play in reporting a person s impres-
sions of numerosity. RAFAEL E. NUNEZ

Department of Cognitive Science, University of California, San Diego, La
Jolla. CA 92093, USA

References

L 6. La toff. ft. Nunez. Where Mathematics Comes From (Basic Books.
New York. 2000).

2. K. Menninger. Number Words and Number Symbols (MIT Press,
Cambridge. MA 1969).

3. J. Wallis. Treatise on Algebra (1685). chapter LXVI.

Response

WE AGREE WITH NUNEZ THAT THE MUNDU-
rucu do not master the formal properties of
number lines and logarithms, but as the term
“intuition” implies, they spontaneously expe-
rience a logarithmic mapping of number to
space as natural and “feeling right.”

Contrary to Niifiez’s claims, mappings
of numbers onto space are omnipresent in
ancient mathematics. Systems of measure-
ment in which numbers are applied in a sys-
tematic linear manner to measure lengths (as
well as a variety of physical continua) date
back at least to the third or fourth millennia
BCE in Egypt, Mesopotamia, and the Indus
Valley (7). Babylonians were engaged in the
measurement of lengths, as illustrated, for
example, by a clay tablet giving the length of
the diagonal of the square up to the sixth deci-

mal (see photograph) (2); the realization that
the diagonal and the side of the square were
incommensurate led to a major and fruitful
crisis in Pythagorician mathematics. We did
not find a precise date for the introduction of
formalized number lines, but writings from
the 17th cenUiry indicate that mathematicians
had conceptualized number lines by that
time (J). Wallis’s Treatise on Algebra, cited by
Nunez, was written years after the introduction
of coordinate systems in 1637 by Descartes
and Fermat. Wallis only uses a number line
metaphor to set the ground for the notion of
negative numbers, en route to introducing
complex numbers.

The logarithmic scale for mapping num-
bers onto space is highly resistant to change,
as we observed the production of logarithmic
scales even among educated Mundurucus,

who were proficient in Porttiguese. U.S. chil-
dren of European ancestry show a similar
behavior, both with the present task ( 4) and in
a slightly different version in which they were
instructed to divide a rectangle in a given
number of parts, which should have favored a
linear scale (J). This resistance could explain
the late introduction of formal linear number
lines in mathematics, despite a solid intuition
for number-space mapping. In the absence of
a formalized concept of logarithms, a logarith-
mically compressed scale presents limited
computational utility, in contrast to a linear
scale, which embodies properties of addition
and subtraction and can be used as a ruler.
Initially, only linearly scaled number lines
were pursued by mathematicians, yet intro-
ducing these linear scales required overcom-
ing the robust logarithmically shaped intuition

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1293

LETTERS

that numbers should be mapped onto space
based on their ratio similarity. Today, the uni-
versal use of logarithmic axes, even in news-
paper graphs, and the extraordinary spread of
slide rules before the advent of calculators
testify to the status of the logarithm as both an
abstract mathematical concept and an intu-
itive mental tool. veronique IZARD, 1 *** 4
STANISLAS DEHAENE,*- 2 *- 4
PIERRE PICA, 4 ELIZABETH SPELKE 4
‘INSERM, Cognitive Neuro-imaging Unit, IFR 49, Gif -stir
Yvette, France. 2 CEA, NeuroSpin center, IFR 49, Gif-sur
Yvette, France. ‘Universite Paris-Sud, IFR 49, F-91191 Gif
sur-YVette. France. ‘Department of Psychology, Harvard
University, Cambridge, MA 02138, USA. 'College de
France, Paris, France. ‘Unit6 Mixte de Recherche 7023,
Formal Structure of Language, CNRS and Paris VIII
University, Paris, France.

References and Notes

1. H. A. Klein, The World of Measurements Masterpieces.
Mysteries and Muddles of Metrology (Simon & Schuster.
New York, 1974).

2. Images of a clay tablet from the Yale Babylonian collec-
tion (MVMr.math.ubc.ca/people7faculty/casVEuctid/ybc/
ybc.html).

3. R. Descartes, la Geometric, D. t Smith and M. L
lantham. TransL. Eds. (Dover. New York. 1954).

4. R. S. Siegler. J. E. Opier. Psych. See 14. 237 (2003).

5. P. Grko. A Mori, Structures Numeriques Etementaires:
Etudes d'Epistemotogie Genetique (Presses Universitaires
de France. Paris. 1962). vol 13.

The Risks of Pigging Out
on Antibiotics

THE NEWS STORY "THE BACTERIA FIGHT BACK"
by G. Taubes (Special Section on Drug Resis-
tance, 18 July, p. 356) highlights the growing
health threat from antibiotic-resistant bacteria,
especially methicillin-resistant Staphylococ
cus aureus (MRSA), and the need to rein in
medical uses of antibiotics to curb resistance.

But reining in health care uses alone is
insufficient to address the resistance epi-
demic. As recommended by the Institute of
Medicine (7), World Health Organization (2),
American Academy of Pediatrics (3), and
other health organizations, routine and wide-
spread use of medically important antibiotics
in animal agriculture also must be ended to
effectively address resistance. Recent evi-
dence showing that some human MRSA
infections are associated with animal agricul-
ture underscores this point.

In Europe, MRSA has been shown to be
transmitted from pigs to farmers and their
families, veterinarians, and hospital staff
(4, 5). One MRSA strain, once found only in
pigs, is associated with serious human illness,
including skin, wound, lung, and heart infec-
tions (6, 7). This new pig strain is linked to
more than 20% of human MRSA infections
in die Netherlands (#).

Researchers have only begun to examine

MRSA from North American livestock. Both
Canadian pig farmers and swine are com-
monly colonized by MRSA (9). A recent
study found that 70% of the tested pigs in
Iowa and Illinois carried MRSA (JO).

Extensive use of antibiotics in livestock
operations can select for resistant bacteria
such as MRSA, just as in health care settings.
By one estimate, more than 70% of all anti-
biotics and related drugs used in the United
States are used as feed additives for livestock
(JJ). Dutch pig farms that routinely use
antibiotics are more likely to have MRSA than
farms with limited antibiotic use (12).

According to the World Health Organi-
zation, “Our grandparents lived during an age
without antibiotics. So could many of our
grandchildren” (13). Overuse of antibiotics in
agriculture as well as in human medicine
could result in this frightening outcome.

REBECCA GOLDBURG,’* STEVEN ROACH, 2
DAVID WALLINGA, 5 MARGARET MELLON 4
‘Environmental Defense Fund, New York, NY 10010, USA.
‘Food Animals Concerns Trust, Chicago, IL 60614, USA.
‘Institute for Agriculture and Trade Policy, Minneapolis, MN
55404, USA. ‘Union of Concerned Scientists, Washington,
DC 20006, USA

*To whom correspondence should be addressed. E-mail:
bgoldburg@edf.org

References

1. M. S. Smolinsld. M. A. Hamburg. J. lederfcecg. Eds..
Microbial Threats to Health: Emergence, Detection, and
Response (National Academy of Sciences, Washington.

DC. 2003).

2. World Health Organization. WHO Qobat Principles for
the Containment of Antimicrobial Resistance in Animals
Intended for Food (WHO Publication WHO/CDS/
CSR/APH/2 000.4. 2000); bttp:/tahqlibdoc.who.int/hq/
200Q/who_cds_csr_aph_2000. 4.pdf.

3. K. M. Shea. Pediatrics 114, 862 (2004).

4. X. W. Huijsdens etal., Ann. Clin. Microbiol. Antimicrob.

5. 26 (2006).

5. A Voss. F. Loeffen. ). Bakker. L. Klaasen. M. Wulf.
Emerging Infect. Dis. 11, 1965(2005).

6. M. 8. Ekkelenkamp. M. Sekkat N. Carpaij. A Troelstra,

M. J. Borneo. Ned. Tijdschr. Geneeskd. 150. 2442 (2006).

7. W. Witte, B. Strommenger. C. Stanek. C. Cuny. Emerging
Infect. Dis. 13. 255 (2007).

8. I. van loo etal.. Emerging Infect Dis. 13. 1834 (2007).

9. T. Khanna. R. Friendship. C. Dewey, J. S. Weese. Vet
Microbiol. 128. 298 (2007).

10. T. C. Smith et at., paper presented at the 2008
International Conference on Emerging Infectious
Diseases, Centers for Disease Control and Prevention,

Letters to the Editor

Letters (-300 words) discuss material published
in Science in the previous 3 months or issues ol
general interest. They can be submitted through
the Web (www.submit2science.org) or by regular
mail (1200 New York Ave., NW, Washington, DC
20005, USA), letters are not acknowledged upon
receipt, nor are authors generally consulted before
publication. Whether published in full or in part,
letters are subject to editing for clarity and space.

Council of State and Territorial Epidemiologists, Atlanta.
GA. 16 to 19 March 2008.

11 M. Mellon. C. Benbrook. K. L Benbrook. Hogging Itl
Estimates of Antimicrobial Abuse in livestock (UCS
Publishing, Cambridge, MA. 2001).

12. E. van Duijkeren etal.. Wl. Microbiol. 126, 383 (2007).

13. World Health Organization. World Health Organization
Report on Infectious Diseases 2000: Overcoming
Antimicrobial Resistance (WHO Publication, 2000);
www.who.int/infeaious-disease-report/2O0OAndex.html

Battle of the Bugs

IN THE NEWS STORY "THE BACTERIA FIGHT
back” (Special Section on Drug Resistance,
1 8 July, p. 356), G. Taubes describes the ongo-
ing war between bacteria and antibiotics,
which the bacteria appear to be winning.
Against this backdrop, scientists are strug-
gling to uncover viable therapeutic alterna-
tives to these erstwhile wonder drugs.

One such alternative, probiotic therapy,
has become the focus of considerable research
effort in recent times (7). Indeed, several clin-
ical trials have attributed impressive health-
promoting effects to probiotics — so-called
“good bugs” — including effective antagonis-
tic activities against a variety of microbial
pathogens by competitive exclusion and bac-
teriocin production (2).

Furthermore, a new generation of probi-
otics termed “designer probiotics” has been
engineered to express proteins that mimic cell
surface receptors, which adsorb toxins and
specifically target enteric infections by block-
ing ligand-receptor interactions between
pathogen and host cells (3). Blocking bacterial
adherence reduces infection, while toxin neu-
tralization ameliorates symptoms until the
pathogen is eventually overcome by the
immune system.

Indeed, McFarland (4), in her seminal
review on the control of antibiotic-resistant
Clostridium difficile, proposed that effective
treatment must “reduce the burden of C. diffi-
cile and its toxins in the intestine, restore the
normal colonic microflora and assist the host’s
immune system.” Designer probiotics, satisfy-
ing all of these criteria, provide an ideal alter-
native for the treatment of not only C. difficile
but also other multidrug-resistant pathogens.

Perhaps the only hope of winning the
war against “bad bugs” will be achieved by
recruiting “good bugs” as our allies.

ROY D. SLEATOR* AND COLIN HILL
Alimentary Pharmabiotic Centre, University College Cork,
College Road, Cork, Ireland.

*To whom correspondence should be addressed. E-mail:
r.sleator@ucc.ie

References

1 IL D. Sleator, C. Hill lett AppL Microbiol. 46. 143 (2008).

2. $. C. Core etal.. Proc. Noll Acad. So. USA. 104, 7617

(2007).

1294

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

LETTERS

3. A. W. Paton. R. Morona. J. C. Paton. Not. Rev. Microbiol.

4. 193(2006).

4. L V. McFariand. J. Med. Microbiol. 54. 101 (200S).

DOE Should Keep
Education in Mind

I BELIEVE A. CHO'S NEWS OF THE WEEK STORY
“Two U.S. labs vie for long-delayed exotic
nuclei source” (18 July, p. 328) misses a few
important points about the competition
between Michigan State University (MSU)
and Argonne National Laboratory (ANL) to
build the Department of Energy-funded
Facility for Rare Isotope Beams (FRIB). I
know both institutions, having received a
PhD. in nuclear physics from MSU in 1967
and having worked as a contractor during the

1970s at ANL, where I taught courses on the
assay of nuclear materials to members of the
International Atomic Energy Commission.

Cho frames this issue as a David-versus-
Goliath contest, comparing the size of
MSU’s existing nuclear science facility,
the National Superconducting Cyclotron
Laboratory (NSCL), with ANL in its en-
tirety. A more apt, apples-to-apples compar-
ison would have considered the relative
scale of MSU ($1.7 billion annual budget,
1 1 ,700 employees) and ANL ($530 million
annual budget. 2,800 employees) (1).
Clearly, both institutions are capable of
managing large, complex operations. And
of the two, only MSU, the nation’s eighth
largest university, is host to a lab designated
as one of the nation’s flagship nuclear

CORRECTIONS AND CLARIFICATIONS

News of the Week: 'Full-genome sequencing paved the way from spores to a suspect" by M. Enserink (15 August, p. 898)
the affiliation of microbial genomicist Claire Fraser -Liggett should have been the University of Maryland School of Medicine
in Baltimore.

Table of Contents: (25 July, p. 457). In the description of the Report 'Did cooling oceans trigger Ordovician biodiversifi
cation? Evidence from conodont thermometry" by J. A. trotter el al., the time of the cooling trend through the Early
Ordovician was incorrect. The sentence should read: 'About 470 million years ago, ocean temperatures dropped to values
near those of today after being much higher for many millions of years, coeval with a sharp jump in biodiversity."
Reports: 'Did cooling oceans trigger Ordovician biodiversification? Evidence from conodont thermometry" by J. A. Trotter
et at. (25 July, p. 550). In the References and Notes, reference 27, which is a duplicate of reference 4, should have been
deleted. References and notes 28 to 39 should have been renumbered 27 to 38. The citations to these references are cor-
rect. In the Fig. 3 legend, the citation to Chen et al. should be 29 (not 31).

Perspectives: 'Tracking corrosion cracking" by A. Stierie (18 July, p. 349). The last sentence on p. 349 incorrectly stated
that ‘King et al. have found that the grain boundaries in stainless steel ... are more sensitive to carbon segregation and
the formation of chromium carbides, which makes them more sensitive to corrosion." Instead, the sentence should state
that ‘King et al. have found that the grain boundaries in stainless steel ... are more sensitive to corrosion, which might be
related to enhanced carbon segregation at these grain boundaries and the formation of chromium carbides."

Reviews: 'Rise of the Andes" by C. N.Garzionee(o/.<6 June, p. 1304). A minus sign was missing from an equation in the
second paragraph of the third column on p. 1305. The correct equation should read “h - -472.5S le O (tln(jU - 2645."
Reports: ‘Hidden neotropical diversity: Greater than the sum of its parts" by M. A. Condon et al. (16 May, p. 928). The first
sentence of the main text included a misplaced reference citation. The sentence should read: "The diversity of neotropical
herbivorous insects, ranging in number from 3 million to 30 million species (1), has been hypothesized to be a function of
plant diversity (2), but the degree to which specialization shapes that function is contentious (3).'

TECHNICAL COMMENT ABSTRACTS

Comment on “Fire-Derived Charcoal Causes Loss of Forest Humus"

Johannes Lehmann and Saran Sohi

Wardle et al. (Brevia, 2 May 2008, p. 629) reported that fire-derived charcoal can promote lossof forest humusand
belowground carbon (0. However, C loss from charcoal-humus mixtures can be explained not only by accelerated
lossof humus but also by lossof charcoal. It is also unclear whether such loss is related to mineralization to carbon
dioxide or to physical export.

Full text at www.sciencemag.org/cgi/content/full/321/5894/1295c

Response to Comment on "Fire-Derived Charcoal Causes Loss of Forest Humus"

David A. Wardle, Marie-Charlotte Nilsson, Olle Zackrisson

We find the suggestion that substantial charcoal loss occurred in the humus-charcoal mixtures implausible and dis-
cuss why compiexing of soluble carbon released from the mixtures by underlying mineral soil should be minor. This
exchange highlights our limited knowledge about charcoal effects on native soil carbon, indicating that strong
advocacy for charcoal addition to offset CO ; emissions remains premature.

Full text at www.sciencemag.org/cgi/content/full/321/5894/1295d

physics facilities by the Nuclear Science
Advisory Committee (2).

However, size is not the critical issue in
this competition. Although the national labs
provide many training opportunities, educa-
tion is of primary importance only in a uni-
versity setting. MSU has the second-best U.S.
nuclear physics graduate program (behind
MIT) and trains 10% of the nation’s nuclear
physics Ph.D.’s (3). The MSU lab is currently
training about 100 graduate and undergradu-
ate students.

Cho points out that the MSU lab might
close if FREB lands elsewhere. So, as the
competition proceeds, decision-makers must
consider whether expanding our national lab
complex at the expense of jeopardizing a
successful university -based educational and
scientific center is in the national interest.

LORENZ A. KULL

Former President and Chief Operating Officer, Science
Applications international Corporation (SAlC), 274 Pearl
Lane, Silverthome, CO 8C498, USA.

References

1. 2007 Michigan State University Data Digest p. 3
(http://opbweb.opb.msu.edu/).

2. The Frontiers of Nudear Science: A Long Range Plan.
DOE/NSF Nudear Science Advisory Committee. 2007,
p. 3 (www.scdoe.gov/hp/nsac/docs/Nudear-Srience.
Low-Res.pdf).

3. U.S. News 8 Woild Report 'America's Best Graduate
Schools.' 2009 edition <http://grad-schools.usnews.
rankingsandreviews.com/gradjphy/nuclear).

Call for an Objective
DOE Decision

AFTER READING A. CHO'S NEWS OF THE WEEK
story “Two U.S. labs vie for long-delayed
exotic nuclei source” (18 July, p. 328), I can’t
help but wonder How can anyone be sure that
the U.S. Department of Energy (DOE) will
make this decision objectively? Our National
Laboratories have outlived the reasons for
which they were established, and, as would
any large and aging organization, they are
vying to f nd a new raison d’etre.

Our universities are hurting for American
students, particularly in technical fields.
Universities, not national labs, have been
a fountain of fresh, competent, and cost-
effective personnel that will provide leader-
ship and allow us to regain our technological
edge in the coming decades. Support for
universities is sorely needed at this time, and
DOE must bear this in mind when they make
their decision, particularly when the facts
point them in a direction that is against their
own interests.

CONSTANTINE CASSAPAKIS

President and CEO, L'Garde, Inc., 15181 Woodlawn Avenue,
Tustin.CA 92780, USA.

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1295

PUBLIC HEALTH

The Tobacco Strategy Entrenched

Carl F. Cranor

E ver wonder why it has been so slow and
difficult to reduce the health risks from
tobacco, secondhand smoke, lead,
beryllium, or chromium? David Michaels's
excellent Doubt Is Their Product provides part
of the explanation, showing
numerous ways in which “the
product defense industry” uses
scientific (and pseudoscien-
tific) arguments to undermine
public health protections, cor-
rupt the scientific record, and
mislead the public.

The book’s title announces
its central theme. A tobacco
industry strategy memo argues,

“Doubt is our product since it is
the best means of competing
with the ‘body of fact' that exists in the minds
of the general public. It is also the means of
establishing a controversy.” (7) The aim: to sow
doubt in the minds of the public, judges, and
even regulatory scientists (if they are suscepti-
ble) about the scientific basis for greater public
health or environmental protections (think
global warming) or tort law actions. Because of
the tobacco industry's success in obftiscating.
slowing, reducing, and blocking regulatory
actions, its approach lias been adopted by oth-
ers, has become institutionalized in presidential
administrations, and has been used as talking
points by some politicians. Fostering doubt and
controversy and demanding high degrees of
certainty postpone legal actions, keep products
in commerce longer, and perhaps delay im-
proved protections indefinitely. They can also
leave the public or work force at risk.

As Michaels (an epidemiologist at George
Washington University) explains, this clever
strategy permits people to oppose public
health rules without arguing the policy point
and without being labeled anti-public health.
It also uses a common science term that might
resonate with some in the scientific commu-
nity. Scientif c articles usually note uncertain-
ties about the research subject and the need for
further studies.

Industries and their supporters have also
demanded “proof” (more at home in mathe-

The reviewer, the author of Toxic Torts: Science, Law, and
the Possibility of Justice, is at the Department of
Philosophy, University of California Riverside, 9C0
University Avenue, Riverside, CA 92521, USA. E-mail-
car l.cranor@ucr.edu

matics than science) before agencies can
increase public health protections or plaintiffs
can successfully receive tort law compensa-
tion for injuries suffered. However, even
though scientists may not understand all
aspects of a problem, public
health agencies need to act on
the weight of the best science
available at the time.

The doubt strategy is most
at home in postmarket legal
contexts. Public health agen-
cies face the burden of estab-
lishing scientific and legal
cases that will withstand ap-
pellate court scrutiny before
they can successfully provide
increased health and environ-
mental protections or withdraw drugs or pesti-
cides from the market. Companies that em-
phasize scientific uncertainties appear to be
scientific angels; they only seek to preserve
the integrity and certainty of the relevant
fi elds against hasty regulatory action based on
incomplete science.

Tort plaintiffs face similar burdens. They
must show that the defendant's products or
actions more likely than not can and did cause
injuries from which the plaintiff suffers.
The U.S. Supreme Court’s Daubert decision

requires judges to review the scientific basis
of expert testimony; they may bar litigants
from trial if the science is insufficient (2).
Fostering doubt may sway judges, who are
typically less informed about science than are
regulatory scientists. The product defense
industry has also helped to persuade some
judges that they should review and exclude
scientific studies individually without review-
ing the total body of relevant evidence on
which scientists rely, a most unscientific way
to review the basis of expert testimony.

The book presents examples of product
defense experts who have accepted binding to
reach predetermined conclusions, misrepre-
sented scientific claims, hidden their affilia-
tions, written articles while using others’
names, or had scientific papers ghost-written
by lawyers.

What should be done? Among the author’s
recommendations is to require the testing of
chemicals before workers and the public are
exposed. If all products were subject to pre-
market testing for safety and impartial agency
review before commercialization, this re-
moves some incentives to raise doubt about
the science. Drug and pesticide manufacturers
rarely point out that their science is too uncer-
tain to permit their products into the market.
Michaels might have said more about addi-
tional legal changes that would reduce the
influence of the doubt and uncertainty argu-
ments, e.g., shifting legal burdens to the man-
ufacturer once its product’s safety was called
into questioa

Michaels also recommends a number of
disclosures: of any and all research sponsors,

Doubt Is Their Product

How Industry's Assault
on Science Threatens
Your Health
by David Michaels
Oxford University Press,
Oxford, 2008. 384 pp.
S27.95, £14.99.

ISBN 9780195300673.

1296

5 SEPTEMBER 2006 VOL 321 SCIENCE www.sciencemag.org

BOOKS ETAL

of what manufacturers know about the toxic-
ity of their product (with penalties for cover-
ing up or lying), and of hazards in the public's
midst (like community right-to-know laws).
These recommendations are not panaceas but
make good f urst steps.

In addition. Doubt Is Their Product re-
minds one of deeper risks that threaten sci-
entific fields and democratic deliberation.
When science affects commercial interests,
there are substantial temptations for re-

searchers or their employers to substitute the
ethics of the marketplace for the ethics of
careful, objective evaluation of the data to
understand the world, environmental threats,
and health risks. Such substitution can result
in the corruption of the scientific literature
and the breaking of incremental links in
chains of evidence on which researchers and
the public depend, and it also tends to under-
mine properly informed political and judicial
decisions. The scientific community and the

public need to be on guard against such
abuses; Michaels’s history of these events
sounds an alert that must not be ignored.

References

L Brown and Williamson Tobacco Company. Smoking and
Health Proposal (Brown and Williamson document no.
<80561778-1786. 1969);
h ttp://leg acy. library, ucsf.edu/ti d/nvs40f00.

2. Daubed v. Merrell Doer Pharm.. Inc . , 509 U.S. 579
(1993).

10. 1126/sciertce. 1162339

THEGONZO SCIENTIST

Chasing the Biggest Shadow of All

Choosing which extreme sport to pursue in one's life is difficult. Most
people are content with the likes of bungee jumping, ice climbing, or
street luge, but not scientists. In addition to thrills, they want their sport
to produce useful data. I tried out an extreme
I* scientific sport last month: eclipse chasing. The

U PI 1 1 PI G objective is to take very sensitive equipment to
sciencemag.org wr y remote locations ' ver Y punctually.

0 For more on this The roots of the s P ort 9° back t0 ancient
episode, go to China, where astronomers experienced, in the
www.gonzosdentist.org words of the late television anchorman Jim
McKay, both "the thrill of victory" (prestige
in the emperor's court) and "the agony
of defeat" (beheading for miscalculation).

Eclipse chasing has come a tong way since
then (more data, less beheading) — and
made headlines around the world in 1919.

On 29 May that year, after struggling with
biting insects and tropical storms on a
volcanic island, a British team recorded
starlight bent around the eclipsed Sun by
gravity, an observation that was widely
trumpeted as confirming Albert Einstein's
theory of relativity.

For my first taste of eclipse chasing, I
joined a team of scientists (1) hoping for a
rendezvous with an eclipse 1 August in the
wild west of Mongolia. To get to the site, we
made a night trek over the Altai Mountains,
which nearly killed us when our driver nod-
ded off at the wheel. On the day itself, we worked in the intense heat and
dust of the Gobi desert, which actually did kill a telescope motor and cam-
era. But just as the eclipse was getting started, I drove a few kilometers
away with astrophysicist Ray Jayawardhana, to take part in a shamanistic
ritual that involved a hundreds-strong chorus of screaming, shouting,
and clapping at the sky. We found ourselves surrounded by terrified
Mongolian locals convinced that a monstrous god called Rah was eating
< the Sun. But that is another story.

$ While Rah captured the Sun, our team captured gigabytes of data. Like
| astro-paparazzi, we harvested hundreds of digital images through a pair of
| telescopes — a refractor and a reflector fixed to a motor-driven astrograph

| built by team member Kosmas Gazeas — during the 2 hours of partial and
2 minutes of total eclipse. And we weren't the only ones ogling the dark-

ened sky. A team led by Jay Pasachoff, an astrophysicist at Williams
College, Massachusetts, was observing to the north in Siberia (2). And to
the south. Science's Beijing correspondent. Richard Stone, was watching
in western China with researchers from the National Astronomical
Observatories, Chinese Academy of Sciences, and other institutions (3).

But how useful are all those data? With orbiting telescopes like Hinode
trained on the Sun — and capable of creating their own eclipse anytime by
simply occluding the Sun's photosphere with a metal disk, can ground-
based observation add anything? "I get that question all the time," com-
ments Pasachoff. In fact, he says, data produced by earthly eclipse chasers
are more valuable than ever. The space telescopes, put in place at enor-
mous cost, provide only part of the picture. By design, "the spacecraft
can't observe a huge region around the Sun, the whole inner and middle
corona." Studying the dynamics of these
superhot solar gases should lead to better
modeling of solar wind and answer a nagging
riddle: Why is the corona hundreds of times
hotter than the Sun's surface? Not only are the
eclipse chasers equipped with "more modern
and efficient" charge-coupled device cam-
eras, explains Pasachoff, but "the resolution on
the corona that we get by processing eclipse
images is finer than that obtainable by any
spacecraft." To understand the Sun, astron-
omers still need the Moon to cover it.

Pasachoff, who has seen 47 solar eclipses,
wants to rename the sport. Rather than a
chaser, "I am an eclipse preceder," he says.
After all, successfully predicting and getting to
the site of an eclipse is the name of the game.
For next year's eclipse, a blockbuster event in
the International Year of Astronomy (4), record numbers of people are
expected to chase — or rather, precede — the 22 July solar eclipse in Asia.
So prepare your telescopes and book your tickets now.

References and Notes

1. The team comprised four astrophysicists— Ray Jayawardhana (University of Toronto),
Kosmas Gazeas (Harvard-Smithsonian Center for Astrophysics), Kazuhiro Sekigochi
(National Astronomical Observatory, Japan), and Katrien Kolenberg (University of
Vienna) — and remote sensing researcher Tuvjargal Norovsambuu (National University
of Mongolia).

2. www.williams.edu/astronomy/eclipse/eclipse20O8.

3. R. Stone. Science 321 . 759 (2008).

4. www.astronomy2009.org.

-JOHN BOHANNON

10.1126/science. 1164877

www.sciencema 9 .org SCIENCE VOL 321 5 SEPTEMBER 2008

1297

MEDICINE

Life Cycle of Translational Research
for Medical Interventions

Despina G. Contopoulos-loannidis . 1 George A. Alexiou , 2 Theodore C. Gouvias . 2 John P. A. loannidis 23 *'

D espite a major interest in translational
research {1-3), development of new,
effective medical interventions is dif-
ficult. Of 101 very promising claims of new
discoveries with clear clinical potential that
were made in major basic science journals
between 1979 and 1983, only five resulted in
interventions with licensed clinical use by 2003
and only one had extensive clinical use {4).
Drug discovery faces major challenges (5-5).
Moreover, for several intervention; supported
by high-profile clinical studies, subsequent evi-
dence from larger and/or better studies contra-
dicts their effectiveness or shows smaller bene-
f ts (9). The problem seems to be even greater
for nonrandomized studies (9). Here, we pre-
sent the results of an empirical evaluation of the
life-cycle phases of translational research for
selected medical interventions.

We examined key milestones in the life
cycle of translational research for all the inter-
ventions claimed to be effective in at least one
study that received over 1000 citations in the lit-
erature in 1990-2004, on the basis of the Web
of Science. These are the most-cited papers in
the literature of medical interventions {10).
Because they have received the greatest atten-
tion, they provide easily identifiable scientific
milestones. Citation counts are a widely
accepted coinage of recognition. Of course,
several blockbusters may go through an indus-
trial discovery-testing-production process that
does not involve any particular highly cited
paper in the peer-reviewed literature. In these
cases, it is not as clear-cut to isolate one or a few
studies that are indisputable milestones in the
translational process.

Of 49 articles with >1000 citations, we
excluded articles where the intervention was

"Department of Pediatrics, University of loannina School of
Medicine, loannina, 45110, Greece, and Department of
Pediatrics, George Washington University School of
Medicine and Health Sciences, Washington. DC 20037, USA.

2 Clinical Trials and Evidence-Based Medicine Unit,
Department of Hygiene and Epidemiology, University of
loannina School of Medicine, loannina, 45110, Greece.
"Department of Medicine, Tufts Medical Center, and Institute
for Clinical Research and Health Policy Studies, Tufts
University School of Medicine, Boston, MA 02111, USA.
‘Biomedical Research Institute, Foundation for Research
and Technology-Hellas, loannina 45110, Greece.

•Author for correspondence (at the address in footnote 2).
E-mail jioannid@cc.uoi.gr; johr.pa.ioannidis@gmail.com

ineffective, as well as those assessing manage-
ment strategies rather than specific interven-
tions, and we selected only the earliest article
whenever two or more highly cited studies with
>1000 citations had been published on the
same intervention and indication. Thirty-two
interventions for specific indications were thus
evaluated, and we could place the milestone of
when their first highly cited clinical study was
published showing effectiveness (tables SI and
S2). We considered this an important time point
in the translational process and estimated how
long a time (“translation lag”) it had taken from
the initial discovery of each intervention to
reach that point. Highly cited status does not
necessarily mean that these interventions con-
tinue to be considered as effective as proposed
in the original highly cited papers. By the end of
2006, the effectiveness of 19 interventions had
been replicated by other subsequent studies {n
= 14) or had remained unchallenged {n = 5),
whereas the other 13 had been either contra-
dicted (w = 5) or found to have had initially
stronger effects (n = 8) when larger or better
controlled subsequent studies were performed
(table SI).

Translation Lag

To place each discovery in time, we identified
the year when the earliest journal publication
on preparation, isolation, or synthesis appeared
or the earliest patent was awarded (whichever
occurred fust). Overall, the median translation
lag was 24 years (interquartile range, 14 to 44
years) between first description and earliest
highly cited article (see the chart, page 1 299).
This was longer on average (median 44 versus
17 years) for those interventions that were fully
or partially “refuted” (contradicted or having
initially stronger effects) than for nonrefuted
ones (replicated or remaining unchallenged)
{P = 0.004).

In a secondary analysis, we def ned the time
of discovery as the first description (publica-
tion or awarded patent) of any agent in the
wider intervention class (those with similar
characteristics and mode of action). Early
translational work may be performed with dif-
ferent agents in the same class compared with
those that eventually get translated into postu-
lated high-profile clinical benefits. Analyses
using information on the wider class of agents

From the initial discovery of a medical
intervention to a highly cited article is a
long road, and even this is not the end of
the journey.

showed even longer translation lag, with med-
ian of 27 (interquartile range, 21 to 50) years
and similar prolongations of the translation lag
for refuted interventions.

Among the 1 8 nonrefuted interventions
that had a highly cited randomized trial to sup-
port them, the median translation lag was 16.5
years (range 4 to 50 years) in the main analysis
[22 years (range 6 to 50 years) considering the
wider class]. The fastest successful translation
occurred for indinavir (as part of triple anti-
retroviral therapy) and abciximab, both of
which took only 4 years from their patenting to
the publication of a highly cited randomized
trial. Both of these fast successes involved
multidisciplinary work spanning molecular to
clinical research on protease inhibitors and
integrins, respectively.

We also tried to identify the first published
article that described the use of each interven-
tion in humans and the first published article
that described the use of each intervention in
humans for the specific intervention described
eventually in the highly cited study {11). There
was a very large variability in the timing of the
f rst human study and of the first human study
for the specific indication (see the chart, page
1 299). The range for the time from f rst discov-
ery to f rst human use was 0 to 28 years. The
range for the time from first discovery to first
specific human use was 0 to 221 years.

We observed that most highly cited claims
that were eventually refuted had a very slow
translation history preceding them [e.g.,
flavonoids, vitamin E, and estrogens were
already available for many decades before
observational (nonrandomized) studies
claimed implausibly large survival benefi Ts in
the 1990s]. We conclude that claims for large
benefits from old interventions require extra
caution as they are likely to be exaggerated.
Given the considerable refutation rate of even
the most highly cited interventions, extensive
replication and confirmation of proposed
treatment benefits are indicated. New drug
discovery is probably essential for common
diseases where the existing drug armamen-
tarium has been already extensively screened.
Conversely, for uncommon and neglected
diseases, the existing drug options may
remain largely untested, and old drugs may
find interesting new uses {12-14).

1298

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

POLICYFORUM

ID Intervention (earliest
intervention in same class)

Refuted, non randomized
17 Inhaled nitric oxide
29 Flavonoids

6 Postmenopausal HRT

14 Vitamin E
Refuted, randomized
22 Aspirin

15 Vitamin E

7 Endarterectomy-carotid
9 Angioplasty-coronary

1 Zidovudine

8 Stent-coronary

10 rt-PA (Streptokinase)

26 Ramipril (CaptopriO

11 HA- 1A (Abs to endotoxin)

Nonrefuted, nonrandomized

16 Oral retinoic add

Nonrefuted, randomized

28 Folate

25 Spironolactone

2 Zidovudine

12 Tamoxifen

4 Levamisole with fluorouradl

24 Ribavirin with interferon

12 Captopril

13 Captopril

5 Enalapril (Captopril)

31 Bisoprotol (Metoprolol)

21 Carvedilol (Metoprolol)

27 lovastatin (Mevastatin)

19 Pravastatin (Mevastatin)

20 Pravastatin (Mevastatin)

30 Simvastatin (Mevastatin)

23 Clopidogret (Tidopidine)

18 Abciximab (murine GPIIb/llla Mo Ab)

3 Indinavir in triple therapy (Saquinavir)

X

First description of intervention class

First description of intervention
2 First article about human use
First article on specific human use
" Highly cited article
™ Partially or fully refuted

Milestones for the 32 interventions. First description ol agent in wider class, tan box (when the agent used in the highly cited article is not the same as the first described
in its class); first description, cyan box; first human-use article, green box; first specific-human use article, yellow box; earliest highly cited publication, red box; realiza-
tion of full or partial refutation (for contradicted or initially stronger effects), black box. Whenever two or more milestones coincide in the same year, the respective col-
ors are superimposed on that box. Folate, flavonoids, and vitamin E were already in human use at the time of first description. Extending beyond the illustrated time range
were the first description for nitric oxide in 1772 and its first human use in 1800; and the first description of flavonoids in 1898, aspirin in 1853, and of the wider class
of antiendotoxins in 1896. Details for these interventions can be found in tables SI to S5, listed by the ID number. Ab, antibody; GP, glycoprotein; HA-1A, human IgM
monoclonal antibody against endotoxin A; HRT, hormone replacement therapy; mo Ab, monoclonal antibody; rt-PA, recombinant tissue plasminogen activator.

Recommendations for the Future

Our analysis documents objectively show the
long length of time that passes between dis-
covery and translation. As scientists, we
should convey to our funders and the public
the immense difficulty of the scientific dis-
covery process. Successful translation is
demanding and takes a lot of effort and time
even under the best circumstances; making
unrealistic promises for quick discoveries and
cures may damage the credibility of science
in the eyes of the public. The following are
some recommendations for improving the
system, based on our analyses:

• Discovery of new substances and inter-
ventions remains essential, but proper credit
and incentives should be given to accelerate
the testing of these applications in high-qual-
ity, unbiased clinical research and the replica-

tion of claims for effectiveness.

• Multidisciplinary collaboration with
focused targets and involving both basic and
clinical sciences should be encouraged.

• Proof of effectiveness for new inter-
ventions requires large, robust randomized
clinical trials.

• Translational efforts for common dis-
eases should focus more on novel agents and
new cutting-edge technologies; for these ail-
ments, it is unlikely that genuine major bene-
fits from interventions already known for a
long time have gone unnoticed.

References and Notes

1. E. A. Zerhouni. JAMA 294, 1352 (2005).

2. F. M. Manncola./. TransL Med. 1, 1 (2003).

3. J. P. loannidis../. Traml. Med. 2, 5(2004).

4. D. G. Contopoulos-ioannidis. E. Ntzani. J. P. loannidis.
Am.). Med. 114, 477(2003).

5. P. CuatrecasasJ. Clin. Invest 116 , 2837 (2006).

6. G. Duyk. Science 302, 603 (2003).

7. 8. Booth. R. Zemmtl. A lot. Rev. Drug Disco v 3, 451 (2004).

8. D. G. Hackarn. D. A RedelmeierJ/UW 296, 1731 (2006).

9. J.P. loannidis. JAMA 294 , 218(2005).

10. Methods and details for the collection and analysis of data
are available as supporting material on Science Online
along with its supplementary tables SI to SS. These inter-
ventions included 18 drugs, two monoclonal antibodies,
one hormonal therapy, four vitamins or food products,
and three surgical or device interventions. Three drugs
and one vitamin appear two times eadi in the list as they
were used for two different indications. For more informa-
tion and references, see tables SI and S2.

1L For more details, see Methods in the supporting online
material and tables S3 to SS

12. C. R. Chong. 0. J. Sullivan. Nature 448 , 645 (2007).

13. S. Zhu et at. Nature 417 , 74 (2002).

14. J. 0. Rothstein etoL. Nature 433, 73 (2005).

10.1126/science.ll60622

Supporting Online Material

www.sciencemag.org/cgi/content/full/321/5894/1298/DCl

www.sciencema 9 .org SCIENCE VOL 321 5 SEPTEMBER 2008

1299

GENETICS

Enhancing Gene Regulation

Gregory A. Wray and Courtney C. Babbitt

CENTG2 gene

HACNS1 moncoding sequence

■* 1 GBX2 gene

Human specific substitutions

V 1*1111 Itt II

Human HACNS1 nucleotide sequence

Drosophila cbromoso

-M—

-hh-

N early half a century
has passed since
Francis Jacob and
Jacques Monod demon-
strated that specific noncod-
ing sequences are required to
activate genes that metabo-
lize lactose in the bacterium
Escherichia coli (7). In a pre-
scient observation, they noted
that mutations in these regu-
latory sequences might play
a role in the evolution of
organismal traits. They fur-
ther argued that gene function
is not only based on the bio-
chemical activity of its prod-
uct but also on how the gene's
expression is regulated. This
idea was expanded in 1975 in
an influential paper by Mary-
Claire King and Alan Wilson
(2), who proposed that trait
differences between humans
and chimpanzees are primarily due to regula-
tory changes in gene expression. Decades
elapsed, however, before it was feasible to
begin testing these ideas in detail. Two papers
in this issue, by Prabhakar ei al. on page 1346
(5), and by Hong et al. on page 1314 ( 4 ),
demonstrate the power of combining bioinfor-
matic approaches with experimental tests to
characterize such regulatory regions.

A major impediment to studying the evo-
lutionary importance of mutations in regula-
tory regions is simply knowing where to look.
DNA sequences that regulate the transcription
of genes occupy no fixed position relative to
coding DNA regions and are often diffuse and
widely dispersed. Even when the position of a
regulatory element is known, there is the
added challenge of identifying which muta-
tions have functional consequences. Within
coding sequences, the genetic code imposes
familiar regularities: Mutations that change
protein structure can be identified exhaus-
tively and unambiguously. By contrast, identi-
fying functional mutations within regulatory
regions requires experimental tests of putative
regulatory elements from different species or

Department of Biology and Institute for Genome Science
and Policy, Duke University, Box 90338, Durham, NC
27708, USA. E-mail: gwray@duke.edu

Shadow enhancer Primary enhancer

Binding sites for transcription factors can vary
between two enhancers that regulate the same gene

individuals — a costly and time-consuming
process. Bioinformatic methods offer a way to
identify promising functional noncoding
regions and to narrow the focus for experi-
mental tests.

One approach is to search genomes for
highly conserved blocks of noncoding
sequence (on the assumption that conserva-
tion implies function) and then scan for
instances of rapid sequence divergence on just
one branch of a phytogeny (which implies a
functional change in a single species) (5).
Prabhakar et al. use this approach to identify a
noncoding region they call human-acceler-
ated conserved noncoding sequence 1
( HACNSJ ) (see the figure). To test the func-
tion of this region, they genetically engineered
mouse embryos to express a construct com-
posed of human HACNS1, the promoter ele-
ment of a heat shock gene, and a reporter
gene. Their results show that human HACNS1
drives expression in the mesenchyme of the
early developing forelimb, and later develop-
ing hindlimb, in these mouse embryos. A
comparison of expression patterns driven by
macaque, chimpanzee, and human orthologs
of HACNS1 revealed that consistently strong
forelimb expression is a unique property of
the human version. By testing various combi-
nations of human and chimpanzee HACNSJ

Bioinformatic approaches reveal functional
changes and the evolution of regulatory
sequences that control gene expression.

Identification of enhancer elements. (Top) HACNS1
is a noncoding region of conservation (percent
identity) among eight vertebrate species, with 13
human-specific substitutions (vertical red lines).
HACNS1 drives the expression of a reporter gene
(purple) in the limbs of a developing mouse
embryo. (Bottom) Primary enhancers (red and pur
pie boxes) near a gene are conserved relative to
more distant “shadow" enhancers, which appear to
be less functionally constrained (pink and light pur
pie boxes). These enhancers drive gene expression
(purple) in the Drosophila embryo.

sequences, the authors narrowed down the rel-
evant functional mutations to an 81 -base pair
region containing 13 substitutions that arose
during human evolution. This concentration
of substitutions is highly unusual relative to
the genome as a whole, implying positive
selection on this region during human origins.

What genes does HACNSJ regulate?
This conserved region lies within an intron
of CENTG2, which encodes a guanosine
triphosphatase activating protein that regu-
lates endosomes (membrane- bound vesicles
that transport materials into a cell). It is also
~300 kb downstream of the next nearest gene,
GBX2, which encodes a transcription factor
that is expressed, among other locations, in
developing limbs. If GBX2 is indeed the
target of HACNSJ regulation, the implica-
tions are fascinating. Because of GBX2' s role
in limb development, the authors note
that changes in its expression could have
altered human limb anatomy — producing, for
instance, specializations of the hand that
facilitate tool use, or modifications of the foot
associated with bipedal ism.

Another bioinformatic approach to identi-
fying regulatory elements is to search for clus-
ters of potential transcription factor binding
sites (6). Hong etal. examined data from chro-
matin immunoprecipitation combined with
microarray technology — so-called ChIP-chip
analysis — from the fruit fly ( Drosophila
melanogaster ) genome for taigets of regula-
tion by the transcription factor Dorsal and
known cofactors (4). Surprisingly, they found
that many target genes of Dorsal contain not
one but two clusters of transcription factor
binding sites, implying the presence of multi-
ple regulatory regions with similar function
(see the figure). Although most experimen-
tally verified enhancers lie within a few kilo-
bases of the gene they regulate, some of the

1300

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

PERSPECTIVES

secondary enhancers lie tens of kilobases
away. The authors used transgenic flies to
show that both nearby and more distant
clusters of binding sites drive similar pat-
terns of reporter gene expression for two
genes, brk and sog.

The evolutionary role of these “shadow
enhancers” is unclear. One possibility sug-
gested by Hong et al. is that they provide an
opportunity for natural selection to tinker with
regulatory sequences, while the primary
enhancers maintain essential gene function.
Consistent with this hypothesis, the sequences
of “shadow enhancers” evolve more rapidly
than those of primary enhancers, which sug-
gests that they operate under fewer functional
constraints. The intriguing possibility that
“shadow enhancers” more commonly confer
expression differences among species could
be tested with the type of comparative experi-
mental approach that Prabhakar and col-
leagues applied to HACNSJ.

Although identifying functional changes
in regulatory sequences remains a serious
challenge, these two papers demonstrate the
power of combining bioinformatics and
experimental tests. However, most regulatoiy
elements are neither highly conserved among
species nor composed of clusters of the
same binding motif. Indeed, few of the
well-documented cases connecting noncod-
ing mutations to trait evolution in humans and
flies ( 7-9) involve regions that would have
been identified as functional on the basis of
sequence conservation or motif clustering.

The challenge now is to develop methods
that can recognize functional changes within
a much greater proportion of regulatory
elements. As phylogenetic sampling of se-
quenced genomes grows denser, it is becom-
ing possible to carry out unbiased surveys of
regulatory change based on genome-scale
functional assays (JCf), quantitative genetics
(]]), and tests for positive selection (12).

Decades after Jacob and Monod first specu-
lated about the evolutionary importance of
regulatory mutations, we are in a position to
begin testing their ideas in earnest

References

1. F. Jacob. J. Monod,/ Mol. BioL 3. 318 (1961).

2. M. C. King. A. C. Wilson. Science 188. 107 (1975).

3. 5. Prabhakar et at.. Science 321. 1346 (2008).

4. J.-W. Hong, D. A. Hendrix. M. S. Levine. Science 321.
1314(2008).

5. S. Prabhakar. J. P. Noonan. S. Paabo. E. M. Rubin. Science
314. 786 (2006).

6. M. Markstein, P. Markstein. V. Markstein. FA 5. Levine.
Proc NotL Acad. Sci. USA 99. 763 (2002).

7. N. Compel. B. Prud'homme. P. J. Wittkopp. V. Kassner. S.
B. Carroll. Nature 433. 481 (2005).

8. N. S. tnanah etal.. Nat. Genet 30. 233 (2002).

9. M. T. Hamblin. A. Di Rienzo. Am. }. Hum. Genet 66,

1669 (2000).

10. A. P. Boyle etal.. Cell 132. 311 (2008).

11. E. E. Schadt etal.. Nature 422. 297 (2003).

12. R. Haygood. 0. Fedrigo. B. Hanson. K. 0. Yokoyama.

G. A. Wray. Nat Genet 39. 1140 (2007).

10.112 6/science. 1163 568

ASTRONOMY

The Universe Measured
with a Comb

A technique for wavelength calibration promises
to revolutionize observational astrophysics, in
areas including planet searches and cosmology.

Sebastian Lopez

I n 1962, Allan Sandage predicted that an
expanding universe should cause a drift
in the redshift of cosmological objects,
but noted: “With present optical techniques
there is apparently no hope of detecting such
small changes in redshifts for time intervals
smaller than 10 7 years” (J). Future extremely
large telescopes (with diameters of 30 to 40
m), equipped with powerful spectrometers,
could in principle enable such a measure-
ment. However, measuring a systematic
change in radial velocity of only 1 cm s 1 per
year over the course of about 20 years — a
measurement referred to as the Sandage-
Loeb experiment (2-4 ) — would still be im-
possible if it were not for the recent develop-
ment of a new and exquisite wavelength cal-
ibration technique called “laser frequency
combs” (5). On page 1335 of this issue,
£ Steinmetz et al. apply this technique for the
| f rst time to an astrophysical experiment (6),
> and the results look promising.

I The Doppler effect provides astronomers
£ with a precise method to measure radial veloc-
| ities of stars and galaxies using the observed

<

I Depaitamento de Astronomfa, Universidad de Chile,
8 Casilta 36-D, Santiago, Chile. E-mail: Uapez@das.uchile.cl

shift in wavelength (or frequency) of their
spectral features relative to laboratories on
Earth: The higher the radial velocity, the
stronger the effect. And when the light enter-
ing the spectrometer comes from distant
objects like galaxies or quasars — thus cross-
ing cosmological distances to reach the tele-
scope — their spectra provide information
about the geometry and history of the universe
as a whole.

Because the universe expands, distant
objects can always be assigned with a redshift,
a quantity that cosmologists relate to distance
and time by fitting various parameters to cos-

The basics of a laser frequency comb. A mode-
locked laser creates femtosecond pulses at gigahertz
frequencies, / (top), that are synchronized with an
atomic clock. A spectrum of the pulses (bottom) is
composed of many modes that are uniformly spaced
in wavelength (or frequency) and cover a spectral
bandwidth given roughly by the inverse of the pulse
duration. Each mode's wavelength (or frequency)
does not have to be measured, but instead is given by
a mathematical relation that includes / np , known a
priori with very high accuracy. Laser frequency combs
could therefore become the perfect wavelength cali-
bration technique for astrophysical experiments that
require high accuracy and long-term stability.

mological models. The past decade has seen
a series of breakthroughs in cosmology.
The Wilkinson Microwave Anisotropy Probe
(WMAP) mission delivered images of the
cosmic microwave background that support a

Laser pulses

Wavelength

Spectrum

Bandwidth r 17x

Frequency ►

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1301

PERSPECTIVES

flat universe, and studies measuring distances
to Type la supemovae or the large-scale distri-
bution of galaxies, among others, have estab-
lished that the universe not only expands, but
that the expansion is accelerating — presum-
ably due to the effect of an unknown compo-
nent in the mass- energy budget of the universe
called “dark energy.” Many projects aim to
elucidate what dark energy really is, but all of
them rely on a given cosmological model;
only the Sandage-Loeb experiment could
track the history of the expansion directly,
without any previous assumption on the
geometry of the universe.

What has been the impediment to fully
exploiting present instrument capabilities?
Traditional spectral calibration techniques use
a crowd of emission or absorption lines at
known laboratory wavelengths as reference to
map the detector pixels into wavelengths.
However, calibration units are subject to
uncertainties that unavoidably degrade the
wavelength solution: Lines are not evenly dis-
tributed in the spectral range of interest, have
a wide range of intensities, and sometimes
appear blended. These systematic effects
become the perennial stumbling block for pre-
cision spectroscopy. They limit the capabili-
ties of current high-resolution spectrometers
and hinder experiment repeatability, crucial
for any long-term monitoring.

The recently developed laser frequency
combs (3, 6-9 ) may offer the solutioa Such a
comb is the spectrum of a femtosecond
“mode-locked” laser that delivers pulses at
repetition rates of~l GHz (determined by
the round-trip time in the laser cavity). When
these pulses pass through a spectrometer, a
regular train of modes is produced in the fre-
quency domain, each of them evenly sepa-
rated by (see the figure) and spanning a
spectral bandwidth given by the inverse of the
pulse duration. Because time — and thus fre-
quency — is the most accurately measured
quantity in physics thanks to atomic clocks,
each mode’s frequency (or wavelength) is
accurately known a priori and can be used as a
perfect ruler to calibrate astronomical spectra.

Steinmetz et al. now report the first astro-
nomical spectrum (of the Sun) calibrated with
a laser frequency comb. Besides slightly out-
performing current best standards of accuracy
using just a small bandwidth, the team was
also able to characterize the stability of the
instrument in an unprecedented fashion. Use
of larger bandwidths should allow wave-
lengths to be measured with a stable precision
of 1 part in 10 billion, opening a new era in
astronomical spectroscopy.

Full implementation of laser frequency
combs in large telescopes will require cover-

age of the entire optical range. Once this chal-
lenge is overcome, at least two other astro-
physical experiments besides the future
Sandage-Loeb test should benefit from the
use of this technique.

First, some astronomers have wondered
whether the atomic physics responsible for
the redshifted absorption lines seen in the
spectra of distant quasars has remained the
same over cosmological times. The values
of fundamental constants or combinations
of constants — like the proton-to-electron
mass ratio or the fine-structure constant —
determine the relative positions of the lines
in the quasar spectra. Thus, one could in
principle compare the value of those con-
stants then (“at high redshift”) and now (on
Earth) to determine whether they have
remained constant. By choosing particular
methods and sets of lines, different groups
have arrived at diverging conclusions. After
a decade of research, the debate is now cen-
tered on the systematic effects inherent to
the observations. Laser frequency combs
could help to identify the origin of these
systematic effects.

Second, the precision and stability offered
by laser frequency combs could greatly help

Janet D. Rowley 1 and Thomas Blumenthal 2

S ince the identification of specific
regions in human chromosomes that
undergo recurring structural rearrange-
ments (translocations) and cloning of the
associated breakpoint genes, the fusion of
genes has been viewed as a unique event in
abnormally growing, usually malignant, cells
{1,2). However, the study by Li et al. on page
1357 in this issue (3) indicates quite the con-
trary, turning at least one paradigm of cancer
cytogenetics on its head.

Li et al. report that in normal human
endometrial tissue, there is a low amount of a
messenger RNA (mRNA) that corresponds to
sequences from two genes, JAZF1 on chro-

‘Departmert of Medicine, University of Chicago, 5841
South Maryland Avenue, MC 2115, Chicago, IL 60637.
USA. E-mail; jrowtey@medicine.bsd.uchicago.edu ^Depart
merit of Molecular, Cellular and Developmental Biology,
University of Colorado. Boulder, CO 80309, USA. E-mail;
tom.blumefithal@colorado.edu

astronomers looking for exoplanets (which
orbit stars other than the Sun). Such planets
imprint small changes on the radial velocity of
their solar system, and monitoring the radial
velocities of bright stars has thus become
the most reliable way of finding planets.
However, the smaller the planet, the smaller
the drift in radial velocity. Discovering Earth-
like planets orbiting solar-like stars in the
“habitable zone” (where life could exist)
requires a precision of about 5 cm s 1 and a
stability of about 1 year. This should be an
easy task for this new technique.

References and Notes

1. A. Sand age. Astrophys. ]. 136 . 319 (1962).

2. A. loeb. Astrophys. ). 499. 1111 (1998).

3. M. T. Muiphy et at., Mon. Not. R. Astron. Soc. 380 . 839
(2007).

4 . ]. Liskie et at.. Mon. Not ft Astron. Soc. 386 . 1192
(2008).

5. T. W. HansCh and J. L Hall received the 2005 Nobel Price
in Physics 'for their contributions to the development of
laser based precision spectroscopy, induding the optical
frequency comb technique.'

6. T. Steinmete et at.. Science 321 . 1335 (2008).

7. Th. Ddem. R. Woltzwarth. T. W. Hansch. Nature 416 . 233
( 2002 ).

8. C. Araujo Hauck etat.. Messenger 129. 24 (2007).

9. C.-H li et at.. Nature 452 . 610 (2008).

10.12 6/science. 1163 194

mosome band 7pl5 and JJAZ1/SUZ12 on
chromosome band 17q21. Moreover, this
chimeric mRNA is identical to that seen in
50% ofhuman endometrial stromal sarcomas,
in which there is a 7; 17 chromosomal translo-
cation that results in a gene fusion, even
though no translocation is detected in normal
endometrial cells. The product encoded by the
chimeric mRNA is a fusion protein that is
expressed in cultured cells, and consequently,
could confer cellular resistance to pro-
grammed cell death and increased growth
(under conditions where expression of the
endogenous JJAZ1 gene was suppressed).
Further, this fusion mRNA is expressed in a
cyclical manner in normal endometrial cells,
most readily detected at the beginning and end
of the menstrual cycle when concentrations of
estrogen and progesterone are low'.

How is this fusion mRNA made in the
absence of a corresponding gene fusion?

MEDICINE

The Cart Before the Horse

Chimeric RNAs, transcribed from malignancy-associated chromosomal translocations, can also
arise from RNA trans-splicing in normal cells.

1302

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

PERSPECTIVES

Li et til. propose that the fusion mRNA is
produced by trans-splicing of RNA (see the
figure) in which nucleotides at the 3' end
of JAZFl-e ncoding precursor mRNA are
replaced with those of JJAZ1 -encoding
precursor mRNA. A different form of trans-
splicing is common in several lower animal
phyla ( 4 ), and a few mRNAs have been shown
to be assembled from separate transcripts
in insects (5). And although trans-splicing in
mammalian cells has been reported, the result-
ing chimeric RNAs do not perform obvious
functions and are usually not present in laige
enough amounts to do so.

It has not been clear how two separate
mRNAs are spliced together (<f). Perhaps they
are brought together through the pairing
of nucleotides within noncoding sequences

(introns) between two transcripts, or maybe
each contains a binding site for proteins that
can form dimers or higher-order multimers.
Although sloppiness by thespliceosome — the
cellular machine that removes introns from
precursor RNA — could be an explanation,
only specific pairs of precursor mRNAs
engage in trans-splicing. Perhaps RNAs from
different genes are trans-spliced because they
are transcribed in the same geographic loca-
tion. Alternatively, trans-splicing could occur
more frequently than we realize, but most
cases go undetected.

What makes the study by Li et al. especially
interesting is that trans-splicing is clearly regu-
lated. The mRNA fusion appears only in cells
from endometrial tissue. Its expression is
increased by hormones and hypoxia, with
S much higher expression in late secretory and
e early proliferative stages of the menstrual
> cycle. The key question is whether the chimeric
* RNA is transcribed from some undetected
£ rearranged copies of the two genes. However,
| Li et al. show that there is no such gene
§ rearrangement in cells producing the trans-
| spliced mRNA and that this trans-splicing
u event can be duplicated in vitro. In addition.

they show that a nontransformed human
endometrial stromal cell line had no rearranged
DNA or visually abnormal chromosomes.

Given the absence of any detectable
rearranged DNA in cells producing the
chimeric RNA, the obvious explanation is
rearrangement at the RNA level. To demon-
strate that trans-splicing could account for the
chimera, Li et al. made extracts from a human
endometrial stromal cell line and from a rhe-
sus monkey fibroblast cell line so that they
could detect trans-spliced products by a
sequence difference between the RNA from
the two species. The authors demonstrated in
vitro trans-splicing of the rhes \.\&JAZF1 exons
(coding regions of DNA) to human JJAZ1
exons. Treatment of the rhesus RNA with
deoxyribonuclease (to cleave any DNA that

might be present) did not prevent formation of
the chimeric RNA, confirming that chimeric
RNA arose from trans-splicing.

Is it a coincidence that the same RNA
occurs in normal cells by trans-splicing and
in tumor cells of the same type by DNA
rearrangement? The authors suggest the
intriguing possibility that whatever leads to
the trans-splicing could also lead to the
genomic rearrangement. This could occur by
at least three general mechanisms. The same
sequences could pair at the RNA level to
result in trans-splicing and at the DNA level
to result in genomic rearrangement. How-
ever, sequence analyses of translocation
breakpoints in leukemia reveal large dele-
tions and duplications as well as precise
nucleotide base- pairing (7). Alternatively,
genomic rearrangement could be a direct
result of the trans-splicing event if genes
involved in the rearrangement are brought
into close proximity during the RNA trans-
splicing process. This idea is consistent with
recent reports on the existence of “factories”
for transcription and RNA processing (8-10).
Finally, the RNA created by trans-splicing
could act as a guide RNA to facilitate the

genomic rearrangement, an idea for which
there is precedent (11). In this case, the trans-
spliced RNA would anneal to regions of both
of the chromosomes and guide them in a
DNA recombination event. Indeed, it is possi-
ble that other genomic rearrangements could
be guided by cellular RNAs.

If fusion mRNA is widespread, it could
explain the conundrum that has long per-
plexed cancer geneticists: why fusion mRNAs
can be detected in apparently normal tissues
of healthy people. Such fusions involve
common translocations seen in neoplastic
hematopoietic cells, but never in solid tumors.
If fusion mRNAs are part of normal cell func-
tion. then finding fusions of the immunoglob-
ulin heavy chain gene (IGH) to the BCL2 gene
in normal spleens, which usually reflects the
presence of a t( 1 4; 1 81 translocation in
lymphomas, would not be unexpected
(12). Given that IGH and the MYC
genes frequently colocalize in tran-
scription factories (9), this geography
could provide a mechanism for having
nascent RNAs in juxtaposition; more-
over, the genes themselves would be
close together. Translocations involv-
ing the IGH and IGK/L genes and the
genes encoding T cell receptors
(TCRs) in lymphoid malignancies are
exceptions in that they do not lead to a
fusion mRNA, but rather to altered
regulation of the apparently normal
target protein (13). Presumably all
translocations are mediated by DNA recombi-
nation enzymes, but could this process be
guided by RNA produced by trans-splicing?
The study by Li et al. also raises questions rel-
evant to clinical practice. Potent therapies tar-
geting fusion mRNA and proteins may disrupt
critical pathways of normal cell function.
Increasingly sensitive methods to determine
the presence of a few translocation-bearing
cells leads one to question whether transloca-
tions or normal cell products are being
detected. This is a critical issue because the
search for minimal residual disease is in high
gear, especially for chronic myeloid leukemia
that responds to the drug imatinib, as a “cure”
seems within reach (14). Many patients suf-
fering from this cancer are translocation-neg-
ative on standard cytogenetic analysis, but
show a gene fusion (BCR-ABL) by reverse
transcriptase polymerase chain reaction. For
these patients, especially the ones with very
low amounts of fusion, it is unclear whether
what is being detected is a malignant cell or a
trans-splicing event.

As the search for fusions in normal cells
will likely be fast-paced for the next few
years, two points should be considered, given

Chromosome translocation

/'"Gene A
f transcription

Gene B
transcription

i

Trans-splking

r

Protein A

I

Q^)

Protein B

r

T

| FUSION BIAS |

9

A/B fusion proteins

Fusion RNAs. Either chromosome translocation or RNA trans-splicing can give rise to fusion mRNAs and proteins.
Some chromosomal translocations produce two hybrid genes that may produce mRNAs containing the 5' end of one
gene and the 3' end of the other. Both may encode fusion proteins. Alternatively, normal mRNAs corresponding to both
genes can recombine by trans-splicing that may produce equivalent fusion mRNAs and proteins. Only one of the pos-
sible fusion mRNAs and proteins is examined by Li etol.

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1303

PERSPECTIVES

the findings of Li et al: cell specificity and
regulation of the trans-splicing event. So
choose the fusion to be investigated, mindful
of these constraints.

References

1. J. D. Rowley. Nat. Rev. Cancer 1. 245 (2001).

2. F. MitcTman, 8. Johansson. F. Mertens, Not. Genet. 36.

331(2004).

3. H. If. J. Wang. G. Mor. J. Sklar, Science 321. 1357
(2008).

4. T. Btumenthal. in Wormbook. The C. elegant Research
Community. Ed.. Wocmbook. 10.1895/vrormbook. 1. 5.1,
wvwi.wonrbook.org.

5. R. Dorn. G. Reuter. A. LoewendorT. Proc NotL Acad. Scr.
U.S.A. 98, 9724 (2001).

6. T. Takahara et al.. Mol Celt. 18. 245 (2005).

7. Y. Zhang. J. D. Rowtey. DNA Repair tAmst). 5. 1282
(2006).

8. S. McCracken ef oL . Nature 385. 357 (1997).

9. C. S. Osborne ef at. RtoS Biol. 5. el92 (2007).

10. J. A. Mitchell P. Fraser. Genes Dev. 22. 20 (2008).

11. M. Nowacki eta!.. Nature 451. 153 (2008).

12. S. Jaru. M. Potter. C S. Rabkin. Genes Chromosome
Cancer 36.211 (2003).

13. T. W. McKeithan. Semin Oncol. 17. 30 (1990).

14. A Hochhaus ef al. Blood 111. 1039 (2008).

10.112 6/science. 1163 791

BIOCHEMISTRY

An Enzyme Assembly Line

Janet L Smith' 2 and David H. Sherman' 3

Fatty acid synthases and related megaenzymes
are highly adaptable to new functions as a
result of their modular architecture.

T he fiuidamental polymers of biology —
proteins, DNA, and RNA — are prod-
ucts of repetitive condensation of sim-
ple amino acid or nucleotide building blocks
and are comparatively easy to assemble.
However, other biomolecules require addi-
tional reactions beyond condensation of
building Nocks. Examples are the fatty acids
and the polyketide and nonribosomal peptide
secondary' metabolites. These molecules are
produced by complex enzyme assembly lines
that include multiple catalytic domains. Two
new crystal structures — one reported recently
(7), the other by Maier et al. on page 13 15 of
this issue (2) — enrich our understanding of
how these mega-enzymes function as effi-
cient factories to produce a remarkable range
of metabolic products.

Maier et al. study the fatty acid synthase
(FAS-I) responsible for de novo fatty acid syn-
thesis in the cytosol of animal cells. FAS-I is
homologous in sequence and architecture
with the very large family of modular polyke-
tide synthases (PKSs), which produce a wide
variety of natural products with potential
medicinal value. In 2006, the authors reported
the structure of FAS-I from a 4.5 A electron
density map, in which most domains could be
assigned but no details were visible (3). Their
new structure provides sufficient detail to
understand the fold and the connectivity of six
of its eight domains; however, the flexibly
tethered acyl carrier protein (ACP) and
thioesterase (TE) domains remain invisible.
The other structure, reported by Tanovic etal.
(7), is of an intact module of a nonribosomal
peptide synthetase (NRPS). Except for the

l Life Sciences Institute, University of Michigan, Ann Arbor,
Ml 48109, USA. 2 Department of Biological Chemistry,
University of Michigan, Ann Arbor, Mi 48109, USA
departments of Medicinal Chemistry, Chemistry, and
Microbiology and Immunology, University of Michigan,
Ann Arbor, Ml 48109, USA. E-mail: janetsmith@umich.
edu; davidhs@umich.edu

Gene duplication + loss ol function + insertion fusion

Proliferation and domain loss

Domain recruitment

Monotunctional ancestors

Canonical PKS extension modules

peptidyl carrier protein (PCP), the primary
NRPS domains are not related to those of the
FAS and PKS systems, but the assembly-line
approach is similar.

All data indicate that the ancestor of FAS-I
was a set of monofunctional enzymes, presum-
ably resembling the dissociated FAS (type II,
FAS- IT) that catalyzes fatty acid biosynthesis in
modem bacteria and plant plastids (see die fig-
ure, panel A). Gene duplication, loss of func-
tion, and gene fusion gave rise to die 270-kD
polypeptide that functions as the homodimeric
FAS-I in mammals (see die figure, panel B). A
different mega-enzyme fusion of monofunc-
tional ancestors evolved in fungi (4). Of these
two assembly-line architectures for fatty acid
synthesis, the mammalian FAS-I proved the
more adaptable, and it now exists not only for
fatty acid biosynthesis but throughout the
eubacterial and fungal world for synthesis of
polyketides. Indeed, high-resolution structures
of PKS components (5, 6) provide critical cor-
roboration for the mammalian FAS structure.

Two key features of the FAS-I architecture
explain its remarkable adaptability. First, the
structure is segregated into two w ings: a selec-
tion/condensing wing for addition of new
building blocks, and a modifying wing for
chemical processing of chain elongation inter-
mediates (see the figure, panel B). The heart
of the assembly line is the condensing wing,
where an acyhransferase (AT) domain selects

Assembly-line proliferation. The dissociated FAS-II
(A) evolved into the homodimeric FAS I (B) (dotted
lines outline disordered parts of the FAS-I structure
(2); lighter shades indicate inactivated DH, KR, and
MT domains). Duplication of an ancestral FAS-I gene,
followed by selective deletion, yielded the canonical
extension modules of modern PKS pathways (0. An
even richer diversity of polyketides arose by domain
recruitment. For example, the CurF protein of
the hybrid PKS/NRPS for curacin A (10) includes
decarboxylase (ECH), cyclopropanase (Cpn), and
NRPS domains (D).

1304

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

PERSPECTIVES

a building block and a ketosynthase (KS)
domain adds it to the growing chain. The
dimeric KS also contributes most of the dimer
contacts in the complex.

The second key architectural feature is an
open and flexible design that is ideal for inser-
tion w deletion of catalytic domains, especially
in the modifying wing. Each two-carbon addi-
tion (via malonate) to a fatty acid chain is fol-
lowed by three reactions — keto reduction
(KR), dehydration (DH), and enoyl reduction
(ER) — carried out in the modifying wing of the
FAS-I. A major source of chemical diversity in
polyketides arises from deletion or inactivation
of one or more of these modifying domains
(see the figure, panel C), providing the chemi-
cal variation that is lacking in fatty acids.

In FAS-I and most fungal PKSs, the assem-
bly line is used for iterative synthesis: Each
enzyme domain performs the same reaction at
each extension step on the growing substrate.
In contrast, in most bacterial PKSs, polyketide
synthesis is sequential: Each extension step is
carried out by an individual FAS-I-like “mod-
ule,” offering the possibility to vary the build-
ing block identity and modification chemistry
at each step. This scheme greatly expands
genetic and protein complexity. Several mod-
ules (up to 20 or more) are required to build a
complex polyketide, and specific interactions
of sequential modules must be faithfully main-
tained by fusion or by docking domains ( 7, 8).

A big surprise of the new FAS structure is
a vestigial methyltransferase (MT) domain at
the periphery of the dimer, following the DH
in the polypeptide sequence. Thus, the mega-

enzyme ancestor of FAS-I appears to have had
a methylation reaction as part of its fatty acid
biosynthetic cycle. Was there a prokaryotic
methyl branched-chain fatty acid, unknown to
us today? The MT domain lost its function in
FAS-I, was deleted from most PKS systems,
but exists in some PKSs as an active methyl-
transferase. And herein lies a conundrum; the
ubiquity of PKS pathways in bacteria and
elsewhere strongly argues that the original
FAS-I evolved in a prokaryote. However,
other than Mycobacterium tuberculosis and
related species that generate unusual fatty
acids, we know of no modem prokaryote that
uses a FAS-I for normal membrane lipid fatty
acid biosynthesis (9).

In many PKS modules, the open FAS-I
architecture has been augmented with a vari-
ety of other catalytic domains, such as S-
acetyltransfer, halogenase, cyclopropanase,
decarboxylase, and even entire NRPS mod-
ules (see the figure, panel D) (JO, 11). The
new structure of the terminal module of the
surfactin NRPS ( 1 ) shows how it, too, is
highly adaptable. Like FAS-I, the NRPS has a
solid platform for condensation, including an
adenylation (A) domain to select the amino
acid building block and a condensation (Q
domain to form a peptide link to the growing
chain ( 1 ). The monomeric C-A didomain
(analogous to KS-AT in the FAS-I condensing
wing) is fused to a PCP and a terminal TE
domain. As in the FAS-I structure, the PCP is
flexibly linked to the synthetase by tethers
long enough for it to deliver substrate to the
active sites of all catalytic domains. Unlike the

MICROBIOLOGY

How to Infect a Mimivirus

Hiroyuki Ogata and Jean-Michel Claverie

T he giant DNA “Mimivirus ” (Acantha
moeba polyphaga mimivirus, or APM)
was initially mistaken for a bacterium,
until La Scola et al. classified it as a virus in
2003 (/). This highly unusual virus has more
genes than many bacteria (2), forms the
most complex known virus particle (5), has
a unique DNA delivery system (4), and
encodes aminoacyl-tRNA synthetases (5),
normally restricted to cellular organisms. As
a possible “missing link” between the cellular

Structural and Genomic information Laboratory, CNRS-
UPR 2589, IFR-88, Unrversite de la Mediterranee, Parc
Scientifique de luminy, FR-13288 Marseille, France.
E-mail; ogata@igs.cnrs-mrs.fr; jean-michel.claverie@
univmed.fr

and the viral world, APM’s discovery revived
theories that link DNA viruses to the emer-
gence of the eukaryotic nucleus (6). Large
viruses closely related to APM are abundant
in the sea (7) and may play important roles in
the geochemical fluxes that regulate Earth’s
climate. La Scola et al. now report in Nature
(8) that the APM family has another unusual
property: It is susceptible to infection by
another virus, named Sputnik (after “travel-
ing companion” in Russian).

Sputnik — a small icosahedral virus with a
DNA genome encoding 2 1 genes — was iso-
lated with a new strain of APM from a cool-
ing tower in Paris. Attempts to culture
Sputnik alone in amoeba cells were not suc-

FAS-I structure, the PCP and TE domains are
well ordered in the NRPS module.

The three assembly line types use homolo-
gous domains (ACP or PCP) to carry the
growing fatty acid, polyketide, or peptide via a
pantetheine-linked thioester. The common
thioester chemistry and the adaptable archi-
tecture have resulted in the proliferation of
hybrid PKS-NRPS and even PKS- FAS-I path-
ways found in phylogenetically diverse bacte-
ria (9, 12). The rich diversity of PKS, NRPS,
and hybrid systems demonstrates that nature
has not employed a Henry Ford-like assembly
line, from which the customer could have any
color car so long as it was black. Rather, we
see a modular assembly line that is easily
copied, modified, and adapted to new func-
tion; this is the secret to its success.

References

L A. Tanovic, S. A. Samel. L-0. Essen, M. A. Marahiel.
Science 321. 659 (2008); published online 26 June
2008 <10.1126/sden«.llS98S0>.

2. T. Maier. M leibundgut N. 8an, Science 321. 1315
<20081.

3. T. Maier. S. Jenni. N. 8an, Science 311. 1258 (2006).

4. S. Jenni. M. Leibundgut. T. Maier. N. Ban. Science 311.
1263 (2006).

5. A. T. Kea tinge-Clay. A M. Stroud. Structure 14. 737 (2006).

6. Y. Tang. C. Y. Kim. 1. 1. Mathews. D. E. Cane, C. Khosla.
hoc. Natl Acad. Sci. U.SA. 103. 11124 (2006).

7. M. Thattsi. Y. Burak. B. I. Shraiman. PLoS Comput Biol.

3. 1827 (2007).

8. J.-P. Nougayrede etot.. Science 313. 848 (2006).

9. R. S. Gokhate. P. Saxena. T. Chopra, D. Mohanty. Not.

Prod. Rep. 24, 267 (2007).

10. Z. Chang etol..]. Not Prod.it. 1356(2004).

11. L. Gu etol.. Science 318, 970 (2007).

12. M. A Fischbach. C. T. Walsh. J. Clardy. Proc. Natl. Acad.
Sd.lLSA 105. 4601(2008).

10.1126/science.ll63785

Large DNA viruses such asthegiant
Mimivirus can be infected by smaller viruses.

cessful. However, when amoebae were inoc-
ulated with the two viruses, both Sputnik and
APM virions multiplied. La Scola et al. (8)
show that Sputnik reproduces in the “virus
factory,” the replication and assembly center
built by APM in amoeba cells during their
lytic infection. The virus factory is a DNA-
rich cytoplasmic compartment that appears 4
hours after APM infection and grows to sev-
eral micrometers in diameter. Sputnik viri-
ons reproduce faster than do APM virions; 6
hours after infection, Sputnik virions start to
emerge from the virus factory, while the new
generation of APM virions only appears after
8 hours. Infection with both viruses de-
creases the yield of infective APM virions

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1305

PERSPECTIVES

E3SES3H

Vertical gene transfer from viral ancestor
Horizontal gene transfer from eukaryotic hosts
Horizontal gene transfer from prokaryotes
Horizontal gene transfer from viruses (by virophages)

and results in “sick” APM virions with aber-
rant morphologies. Sputnik thus behaves as a
true parasite with a detrimental effect on
APM reproduction.

Small viruses requiring other larger
viruses for their reproduction have previously
been documented. These “satellite viruses”
lack essentia] functions for multiplication, for
which they exploit their “helper viruses.” La
Scola etal. (5) argue that Sputnik is more than
a satellite virus, because it uses its partner’s
virus factory and impairs its fitness. They
therefore call Sputnik a “virophage.”

What is the origin of the Sputnik viro-
phage? The authors provide evidence sug-
gesting the existence of related virophages in
the oceans (5). Marine virologists have
reported small viruses occurring with larger
ones in marine protist populations (9, JO).
During recurrent infection of a cell by the two
viruses, one virus may begin to benefit from
the other. Like Sputnik, the small marine
viruses multiply faster than the larger ones. If
the viral genomes can physically interact,
genes can be exchanged, and the two viruses
may evolve into various states of depend-
ency, from mutualisms to parasitism. In this
context, it is worth noting that Sputnik has an
integrase (an enzyme that inserts pieces of
DNA from one DNA molecule into another).
The genome of a marine virus, infecting the
planktonic species Emiliania huxleyi con-
tains a strange 1 76-kb central segment (77):
Genes in this segment lack homologs in other
viruses, but harbor a unique promoter. This
segment is expressed much earlier than the
rest of the viral genome and may be the inte-
grated genome of an unknown virophage.

The genes in giant eukaryotic viruses have
multiple origins (see the figure). The APM
genome contains eukaryotic- or prokaryotic-
like genes. Recent horizontal gene transfers
from its eukaryotic hosts or prokaryotic or-
ganisms partially account for these genes.
However, giant viral genomes also contain
genes that are unique to viruses, the origin of
which is hotly debated (6, 12, 13). Do these
genes originate in vertical gene transfer from a

WWiMI

7

-30% shared with cellular organisms

0 - 10 %

0 - 10 %

-70% unique to viral life (virosphere)

very old viral common ancestor? The small
number of genes shared among modem
viruses argues against this possibility. Viral
genome mosaicism is also suggested by the
occurrence of very similar genes in different
viruses (14). Furthermore, a substantial
amount of horizontal gene transfer may occur
between viruses. The Sputnik virophage now
provides a new potential vehicle for such hor-
izontal gene transfers. In fact, the Sputnik
genome encodes several genes that may origi-
nate in vastly different viruses.

E lectrons possess magnetic behavior
through the quantum mechanical prop-
erty of spin. The magnetic properties of
materials then arise from the collective interac-
tion of electrons on atoms within the crystal.
Below a transition temperature, the electron
spins of normal magnets “freeze” into an
ordered array of magnetic dipoles. Whether the
ordering is ferromagnetic (all the diproles point
in the same direction) or anti ferromagnetic (the
dipoles on adjacent sites point in opposite
directions) is determined by the sign and
strength of the interaction between the elec-
trons. Early theoretical work has indicated a
departure from these ordered states, suggesting
that quantum mechanical fluctuations of the
spin could be so strong that ordering would be
suppressed and the spin ensemble would
remain in a liquid-like state, even down to the

Department of Physics, Massachusetts Institute of Tech-
nology, Cambridge, MAC2139, USA. E-mail: palee@mit.edu

Assessing the proprortions of vertical gene
transfer and virus-virus horizontal gene trans-
fer now appears crucial for understanding the
evolution of giant viruses, refining the con-
cept of virus lineage, and elucidating gene
flow in the virosphere. The unusual features of
the giant Mimivirus rerived the popular, yet
unresolved question: “Are viruses alive?” The
discovery that some of them can get sick adds
a new twist to this old debate.

References

L B.U Scola et al. Science 299. 2033 (2003).

2. D. Raoult etal. Science 306. 1344 (2004).

3. P. Renesto et at. ). ViroL 80. 11678 (2006).

4. N. Zauberman etal. BloS Biot. 6. ell4 (2008).

5. C. Abergel et at.. J. Wot. 81. 12406 (2007).

6. J. M. Gaverie. Genome Biot. 7. 110 (2006).

7. A Monier, J. M. Gavww, H. Ogata, Genome Biol. 9.

R106 (2008).

8. B. La Scola et al. Nature 10.1038/nature07218 (2008).

9. K. Nagasaki, J. Microbiol. 46. 235 (2008).

10. C. P. Brussaard et al. Virology 319. 280 (2004).

11 M. J. Allen etal.]. Virol 80. 7699 (2006).

12. H. Ogata. J M. Claverie. Genome Be% 17. 1353 (2007).

13. E. V. Koonin. W Martin. Trendi Genet 21, 647 (2005).

14. K. Nagasaki etal.. AppL Environ. Microbiol 71. 3599
(2005).

!{!{>> m

Ordered spins. (Left) Neel's picture ol antiferro-
magnet ordering with an alternate spin-up-spin-
down pattern across the lattice. (Right) Quantum
fluctuations lead to mutual spin flips, which Landau
argued would disorder Neel's state.

lowest temperatures. Experimental evidence,
which has until recently remained elusive, is
emerging in favor of this long-predicted state of
quantum matter.

To understand the controversy surround-
ing this exotic quantum spin liquid state, it is
instructive to go back to the description of
anti ferromagnetism. Soon after the invention
of quantum mechanics, Heisenberg prointed
out that electron spins on neighboring atoms
can have short-range interaction due to
quantum mechanical exchange. Louis N6el

Origin of genes in large eukaryotic viruses. The distribution of sequence database matches suggests
diverse origins for the genes of large DNA viruses. Horizontal gene transfer may occur through exposure to
host or prokaryotic DNA. The many genes unique to viruses are vertically or horizontally transferred between
viruses— a process in which the newly discovered virophages may play a key role.

10.112 67«:ience. 1164839

PHYSICS

An End to the Drought of
Quantum Spin Liquids

Patrick A. Lee

After decades of searching, several promising examples of a new quantum state of matter have
now emerged.

1306

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

PERSPECTIVES

showed that a negative exchange interaction
results in an antiparallel arrangement of
neighboring spins, and that at low tempera-
tures the spins order in an alternating up-down
pattern (see the first figure, left panel) (7).
This anti ferromagnetic, or N6el, state was ini-
tially greeted with skepticism — Lev Landau
was among the doubters (2).

Landau did not publish hisobjection, but on
the basis of Friedel’s comments (2) we can sur-
mise that his argument was that the correct
application of quantum mechanics to such a
system would lead to fluctuations (see the first
figure, right panel), which may completely ran-
domize the spin order. It was not until 1949,
when antiferromagnetic order was directly
observed by neutron scattering, that N6el was
fully vindicated. In fact, Noel’s theory was far
more success hi! than he had the right to expect:
For 70 years the Neel state has prevailed.

In 1 973, Philip W. Anderson reasoned that
the best chance to defeat Neel ordering came
in the form of “frustrated” spin systems (3),
with the special example of spins on a trian-
gular lattice. This lattice is “frustrated”
because if an up- and down-spin occupy two
comers of a triangle, the spin on the third cor-
ner does not know which way to point to
obtain the lowest-energy configuration.
Instead of a Neel state, Anderson proposed a
ground state made up of a quantum mechani-
cal superposition of singlet pairs that cover
the lattice. He called this a resonating valence
bond (RVB) state, an explicit example of a
quantum spin liquid. Unfortunately, it was
soon shown that neighboring spins in a trian-
gular lattice manage to order at a 1 20° angle,
and Neel once again won out.

The field of quantum spin liquids lan-
guished until 1987, when high-temperature
superconductivity was discovered. Anderson
pointed out a connection between the RVB
spin liquid and the Cooper pairs of a super-
conductor (4). Attempts to justify the RVB
theory of superconductivity led to rapid devel-
opments of the spin liquid theory. It is now
understood that the spin liquid (defined as
having an odd number of electron spins on
each lattice unit cell) is a new state of matter
f with properties we have never encoimtered
| before. For example, the excited states may be
£ spinons — charge-neutral objects that possess
| magnetic properties. Depending on the type

1 of spin liquid, the spinon may obey Fermi or
| Bose statistics and there may or may not be an
'6 energy gap. Furthermore, these spinons can-
| not live by themselves but are generally
§ accompanied by gauge f elds.just as electrons

2 are always accompanied by electromagnetic
a gauge fields (5). This is a dramatic example of
o emergent phenomena, where new particles

and fields emerge at low-energy scales but are
totally absent in the Hamiltonian that de-
scribes the initial system.

Confirmation of the existence of the spin
liquid state has been elusive, and only
recently have several promising examples sur-
faced. The first is an organic solid called
K-(ET) 2 Cu 2 (CN) 3 (6, 7) in which the active
ingredients are dimers of an organic molecule,
ET [bis< ethylenedithio) - tetrathiafulvalene). A
single electron is localized on each dimer,
which forms layers of approximately triangular
lattices. Despite an exchange energy of -250 K,
no magnetic order was detected down to 30
mK. This material is an insulator but becomes a
superconductor (critical temperature T c = 3.5 K)
and then a metal under pressure. It is believed
that the proximity to an insulator-to-metal tran-
sition implies that the spins interact with a more
complicated Hamiltonian than the Heisenberg
model and allows the spin liquid state to form
(8, 9). Remarkably, the spin susceptibility goes
to a constant at low temperatures and the spe-
cific heat is linear in temperature (7). These
properties are normally associated with metals,
being consequences of the electron Fermi sur-
face. The linear specific heat is particularly
unusual foran insulator that is relatively defect-
free. Furthermore, the ratio of magnetic sus-
ceptibility to the linear temperature coefficient
of the specific heat is close to that of free fermi-

Meeting with frustration. (Top) A Kagome basket.
(Bottom) Structure of ZnCu 3 (OH) 4 Ct 2 ( 10 ) showing
that the Cu ions (blue) occupy a Kagome lattice; 0-H
is red white.

ons. These observat ions strongly suggest that the
excitations are indeed fermionic spinons that
form a Fermi sea, and thus offer strong evidence
for a spin liquid ground state.

Last year an entirely different class of spin
liquid was discovered. It has long been sus-
pected that spins on a Kagome lattice support
a spin liquid ground state. Kagome is the
Japanese name for the weave pattern of a bas-
ket (see the second figure, top panel). The
structure consists of comer-sharing triangles
and is even more frustrated than the triangular
lattice considered by Anderson. Last year saw
the synthesis of such a solid-state Kagome
system: ZnCu 3 (OH) 6 Cl 2 , where a single elec-
tron spin resides on the Cu (10) (see the sec-
ond figure, bottom panel). Although the
exchange energy is -200 K, this material does
not show any magnetic ordering down to mil-
likelvin temperatures. The magnetic excita-
tions are apparently gapless, but unlike the
organic compound, the large specific heat at
low temperatures is sensitive to magnetic
F eld, which suggests that the low-temperature
properties may be dominated by a few percent
of local moment defects.

Spin liquids are not limited to two-
dimensional systems. A newly synthesized
material, Na 4 Ir 3 O g , has Ir ions that form a
three-dimensional network of corner-shar-
ing triangles, termed a hyper-Kagome struc-
ture (77). Despite an exchange energy of
-300 K, no magnetic order was found down
to 1 K and below.

It is an exciting time in the history of anti-
ferromagnetism. After decades of searching,
three examples of the defeat of Neel order by
quantum fluctuations have been discovered
in quick succession. There are good reasons
to believe that fermionic spinons will
emerge as the low-energy excitations, but
more work will be needed to confirm this.
An even more intriguing question is whether
their partner, the emergent gauge field, can
make its presence felt as well. We can be
optimistic that even more exciting discover-
ies lie ahead.

References

1 L. Neel An n. Rhys. 5. 232 <1936).

2. ). Friedet. Rhys. Today 54. 88 (October 2001).

3. P. W. Anderson. Mater. Res. Bull. 8. 153 (1973).

4. P. W. Anderson. Science 23S. 1196 (1987).

5. P. A. Lee eta!.. Nat. Rev. Mod. Rhyt. 78. 17 (2006).

6. Y. Shimizu, H Miyagawa. K. Kanoda. M. Maesato.

G. Saito. Rhys. Rev. Lett. 91. 107001 (2003).

7. S. Yamashita et at.. Nat. Rhys. 4, 459 (2008).

8. 0. Motrunich, Rhys. Rev. B 72. 045105 (2005).

9. S.-S. Lee. P. A. Lee. Rhys. Rev. Lett. 95. 036403 (2005).

10. J. Helton ef at.. Rhys. Rev. tett 98. 107204 (2007).

11 Y. Okamoto et at.. Nat. Rhys. Rev. tett. 99. 137207

(2007).

10.1126/sri ence.1163196

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1307

Flood or Drought: How Do Aerosols
Affect Precipitation?

Daniel Rosenfeld, 1 * Ulrike Lohmann, 2 Graciela B. Raga, 3 Colin D. O'Dowd, 4
Markku Kulmala, 5 Sandro Fuzzi, 6 Anni Reissell, 5 Meinrat O. Andreae 7

Aerosols serve as cloud condensation nuclei (CCN) and thus have a substantial effect on
cloud properties and the initiation of precipitation. Large concentrations of human-made
aerosols have been reported to both decrease and increase rainfall as a result of their radiative
and CCN activities. At one extreme, pristine tropical clouds with low CCN concentrations rain
out too quickly to mature into long-lived clouds. On the other hand, heavily polluted clouds
evaporate much of their water before precipitation can occur, if they can form at all given the
reduced surface heating resulting from the aerosol haze layer. We propose a conceptual model
that explains this apparent dichotomy.

C loud physicists commonly classify the
characteristics of aerosols and clouds
into ‘'maritime’’ and “continental” regimes,
where “continental” has become synonymous
with “aerosol-laden and polluted” Indeed, aero-
sol concentrations in polluted air masses arc
typically one to two orders of magnitude greater
than in pristine oceanic air (Fig. 1) ( 1 ). How-
ever, before humankind started to change the
environment, aerosol concentrations were not
much greater (up to double) over land than
over the oceans (/, 2). Anthropogenic aerosols
alter Earth’s energy budget by scattering and
absorbing the solar radiation that energizes the
formation of clouds (3-5). Because all cloud
droplets must form on preexisting aerosol par
tides that act as cloud condensation nuclei (CCN),
increased aerosols also change the composi-
tion of clouds (i.e., the size distribution of cloud
droplets). This, in turn, determines to a large ex-
tent the precipitation-forming processes.

Precipitation plays a key role in the climate
system About 37% of the energy input to the
atmosphere occurs by release of latent heat
from vapor that condenses into cloud drops
and ice crystals (6). Reevaporation of clouds
consumes back the released heat. When water
is precipitated to the surface, this heat is left in
the atmosphere and becomes available to ener-
gize convection and larger-scale atmospheric
circulation systems.

institute of Earth Sciences, Hebrew University of Jerusa-
lem, Jerusalem 91904, Israel, institute For Atmospheric
and Climate Science, ETH Zurich, 8092 Zurich, Switzerland.
} Universidad Nacional Autdncma de Mexico, Mexico City
04510, Mexico. 4 Schcol of Physics and Centre for Climate
and Air Pollution Studies, Environmental Change institute.
National University of Ireland, Galway, Ireland. Apart-
ment of Physics, University of Helsinki, Post Office Box 64,
Helsinki 00014, Finland. nstituto di Scienze detl’Aimcsfera
e del Clima-CNR, Bologna 40129, Italy. 7 Biogeochemistry
Department, Max Planck Institute for Chemistry, Post Office
Box 3060, D-55020 Mainz, Germany.

•To whom correspondence should be addressed. E-mail:
danielrosenfeld@huji.ac.il

The dominance of anthropogenic aerosols
over much of the land area means that cloud com-
position, precipitation, the hydrological cycle,
and the atmospheric circulation systems are all
affected by both radiative and microphysical im-
pacts of aerosols, and arc likely to be in a differ-
ent state relative to the pre-industrial era.

The Opposing Effects of Aerosols
on Clouds and Precipitation

The radiative effects of aerosols on clouds most-
ly act to suppress precipitation, because they de-
crease the amount of solar radiation that reaches
the land surface, and therefore cause less heat to

be available for evaporating water and energiz-
ing convective rain clouds (7). The fraction of
radiation that is not reflected back to space by
the aerosols is absorbed into the atmosphere,
mainly by carbonaceous aerosols, leading to
heating of the air above the surface. This sta-
bilizes the low atmosphere and suppresses the
generation of convective clouds (5). The warmer
and drier air thus produces circulation systems
that redistribute the remaining precipitation ( 8 , 9).
For example, elevated dry convection was ob-
served to develop from the top of heavy smoke
palls from burning oil wells (JO). Warming of
the lower troposphere by absorbing aerosols
can also strengthen the Asian summer monsoon
circulation and cause a local increase in precipi-
tation, despite the global reduction of evaporation
that compensates for greater radiative heating
by aerosols (11). In the case of bright aerosols
that mainly scatter the radiation back to space,
the consequent surface cooling also can aha
atmospheric circulation systems. It has been
suggested that this mechanism has cooled the
North Atlantic and lienee pushed the Intertropical
Convergence Zone southward, thereby contrib-
uting to the drying in the Sahel (12, 13).

Aerosols also have important microphysical
effects (14). Added CCN slow the conversion of
cloud drops into raindrops by nucleating larger
number concentrations of smaller drops, which
are slower to coalesce into raindrops or rime
onto ice hydrometeors (15, 16). This effect was
shown to shut off precipitation from very shal
low and short-lived clouds, as in the case of

Fig. 1. Relations between observed aerosol optical thickness at 500 nm and CCN concentrations at
supersaturation of 0.4% from studies where these variables have been measured simultaneously, or
where data from nearby sites at comparable times were available. The error bars reflect the variability
of measurements within each study (standard deviations or quartiles). The equation of the regression
line between aerosol optical thickness (y) and CCN a4 Or) is given by the inset expression; R is the correlation
coefficient The aerosols exclude desert dust [Adapted from (I)|

www.sciencema 9 .or 9 SCIENCE VOL 321 5 SEPTEMBER 2008

1309

REVIEW

Growing

Mature

Dissipating

Fig. 2. Evolution of deep convective clouds developing in the pristine
(top) and polluted (bottom) atmosphere. Cloud droplets coalesce into
raindrops that rain out from the pristine clouds. The smaller drops in the
polluted air do not precipitate before reaching the supercooled levels,
where they freeze onto ice precipitation that falls and melts at lower
levels. The additional release of latent heat of freezing aloft and reab-

sorbed heat at lower levels by the melting ice implies greater upward
heat transport for the same amount of surface precipitation in the more
polluted atmosphere. This means consumption of more instability for the
same amount of rainfalL The inevitable result is invigoration of the con-
vective clouds and additional rainfall, despite the slower conversion of
cloud droplets to raindrops { 43 ).

smoke from ship smokestacks in otherwise
pristine clouds over the ocean {17). This created
the expectation that polluted areas would suffer
from reduced rainfall. On the other hand, it was
expected that accelerating the conversion of
cloud water to precipitation (i.e., increasing the
autoconversion rate) by cloud seeding would
enhance rainfall amounts. It turns out, however,
that polluted areas are not generally drier, and
rain enhancement by cloud seeding remains
inconclusive (18, 19).

With the advent of satellite measurements,
it became possible to observe the larger pic-
ture of aerosol effects on clouds and precip-
itation. (We exclude the impacts of ice nuclei
aerosols, which are much less understood than
the effects of CCN aerosols.) Urban and in-
dustrial air pollution plumes were observed to
completely suppress precipitation from 2.5-km

deep clouds over Australia (20). Heavy smoke
from forest fires was observed to suppress rain-
fell from 5-km-decp tropical clouds (21, 22).
The clouds appeared to regain their precipitation
capability when ingesting giant (>1 pm diame-
ter) CCN salt particles from sea spray (23) and
salt playas (24). These observations were the
impetus for the World Meteorological Organi-
zation and the International Union of Geodesy
and Geophysics to mandate an assessment of
aerosol impact on precipitation (19). This report
concluded that “it is difficult to establish clear
causal relationships between aerosols and precip-
itation and to determine the sign of the precipi-
tation change in a climatological sense. Based on
many observations and model simulations the ef-
fects of aerosols on clouds are more clearly un-
derstood (particularly in ice- free clouds); the
effects on precipitation are less clear.”

A recent National Research Council report that
reviewed “radiative forcing of climate change"
(2S) concluded that the concept of radiative
forcing “needs to be extended to account for (1)
the vertical structure of radiative forcing, (2) re-
gional variability in radiative forcing, and (3)
nonradiative forcing." It recommended “to move
beyond simple climate models based entirely
on global mean top of the atmosphere radiative
forcing and incorporate new global and regional
radiative and nonradiative forcing metrics as they
become available.” We propose such a new met-
ric below.

How Gin Slowing the Conversion
of Cloud Droplets to Raindrops
Enhance Rainfall?

A growing body of observations shows that sub-
micrometer CCN aerosols decrease precipitation

1310

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

REVIEW

from shallow clouds (17, 20, 21, 26-28) and
invigorate deep convective rain clouds with
warm (> ~15°C) cloud base (29-33), although
the impact on the overall rainfall amount is not
easily detectable (34, 35). These observations are
supported by a large number of cloud-resolving
model studies (36-43). The simulations also
show that adding giant CCN to polluted clouds
accelerates the autoconvcrsion, mainly through
nucleating large drops that rapidly grow into
precipitation particles by collecting the other
smaller cloud droplets (44). However, the auto-
conversion rate is not restored to that of pristine
clouds (42).

Fundamentally, the amount of precipitation
must balance the amount of evaporation at a
global scale. Therefore, the consequence of aero-
sols suppressing precipitation from shallow
clouds must be an increase in precipitation from
deeper clouds. Such compensation can be ac-
complished not only at the global scale (45) but
also at the cloud scale; that is, the clouds can
grow to heights where aerosols no longer im-
pede precipitation (46). All of this is consistent
with the conceptual model shown in Fig. 2. This
model suggests that slowing the rate of cloud
droplet coalescence into raindrops (i.e., auto-
conversion) delays the precipitation of the cloud
water, so that more water can
ascend to altitudes where the
temperanirc is colder than 0°C.

Even if the total rainfall amount
is not decreased by the increase
in aerosols, delaying the forma-
tion of rain is sufficient to cause
invigoration of cloud dynam-
ics. By not raining early, the
condensed water can form ice
precipitation particles that release
the latent heat of freezing aloft
(6, 29, 30) and reabsorb heat at
lower levels where they melt
after frilling.

The role of ice melting below
the 0°C isotherm level in invig-
oration has been successfully
modeled (47), although models
also predict invigoration through
increased aerosol loads even with-
out ice processes (43). These
model simulations suggest that
the delay of early rain causes
greater amounts of cloud water
and rain intensities later in the
life cycle of the cloud. The en-
hanced evaporative cooling of
the added cloud water; mainly
in die downdrafts, provides part
of the invigoration by the mech-
anism of enhanced cold pools
near the surface that push up-
ward the ambient air. The greater
cooling below and heating above
lead to enhanced upward heat
transport, both in absolute terms

and normalized for the same amount of sur-
face precipitation. The consumption of more
convective available potential energy (CAPE)
for the same rainfall amount would then be con-
verted to an equally greater amount of released
kinetic energy that could invigorate convection
and lead to a greater convective overturning, more
precipitation, and deeper depletion of the static
instability (6). Simulations have shown that greater
heating higher in the troposphere enhances the
atmospheric circulation systems (48).

In clouds with bases near or above the 0°C
isotherm, almost all the condensate freezes, even
if it forms initially as supercooled raindrops in a
low-CCN environment. Moreover, the slowing
of the autoconvcrsion rate by large concentra-
tions of CCN can leave much of the cloud drop-
lets airborne when strong updrafts thrust them
above the homogeneous ice nucleation level of
— 38°C, where they freeze into small ice parti-
cles that have no effective mechanism to coag-
ulate and fall as precipitation. This phenomenon
was observed by aircraft (49) and simulated
for convective storms in west Texas (50) and
the U.S. high plains (51). When the same sim-
ulation (50) was repeated with reduced CCN
concentrations, the calculated rainfall amount
increased substantially. The same model showed

— a Water load — C Unload >0 freeze unload

— b Water unload — d Load >0 freeze unload

Fig. 3. The buoyancy of an unmixed adiabatkally raising air par-
cel. The zero-buoyancy reference is the standard parcel: liquid water
saturation, immediately precipitating all condensates without freez-
ing (vertical line b). Cloud base is at 22°C and 960 hPa. The buoy-
ancy of the following scenarios is shown: (a) suppressing rainfall and
keeping all condensed water load, without freezing; (M precipitat-
ing all condensed water, without freezing; (r) precipitating all con-
densates, with freezing at T < -4°C; (d) Suppressing precipitation until
T = -4°C, and then freezing and precipitating all condensed water
above that temperature. The released static energy (] kg -1 ) with
respect to reference line b is denoted by the numbers.

that adding small CCN aerosols in warm-base
clouds has the opposite effect to that of cold-
base clouds: increasing the precipitation amount
by invigorating the convective overturning, while
keeping the precipitation efficiency (i.e., sur-
face precipitation divided by total cloud con-
densates) lower (52).

The invigoration due to aerosols slowing
the autoconversion can be explained according
to fundamental theoretical considerations of
the pseudo-adiabatic parcel theory (Fig. 3). The
CAPE measures the amount of moist static en-
ergy that is available to drive the convection. Its
value is normally calculated with reference to a
pseudo-adiabatic cloud parcel that rises while
precipitating all its condensate in the form of
rain, even at subfreezing temperatures.

Consider the case of a tropical air parcel that
ascends from sea level with initial conditions
of cloud base pressure of 960 hPa and temper
ature of 22°C. When not allowing precipitation,
all the condensed water remains in the parcel
and requires 415 J kg -1 to rise to the height of
the -4°C isothenn (point d\ in Fig. 3), which is
the highest temperature at which freezing can
practically occur in the atmosphere. Freezing
all the cloud water would warm the air and add
thermal buoyancy by an amount that would
almost exactly balance the condensate load (d 2 ).
When the ice hydrometeors precipitate from a
parcel, it becomes more positively buoyant be-
cause of its reduced weight (d 2 ), so that the re-
leased convective energy at the top of the cloud
(d 4 ) is the largest Specifically, it is greater by
-1000 J kg” 1 relative to the case where cloud
water is precipitated as rain below the -4°C iso-
thenn and as ice above that level (c,). However,
lurther delaying the conversion of cloud water
into precipitation to greater heights above the
0°C level weakens the convection. In the ex-
treme case of extending the suppression from
the -4°C to the -36°C isotherm level (a0, ad-
ditional energy of 727 J kg -1 is invested in lifting
the condensates. There is no effective mecha-
nism for precipitating cloud water that glaciated
homogeneously into small ice particles. This
would prevent the unloading of the parcel, tak-
ing up even more convective energy and lurther
suppressing the convection and the precipita-
tion. In reality, cloud parcels always mix with
the environment, but this applies equally to all
the scenarios in Fig. 3, so that qualitatively the
contrasting aerosol effects remain the same.
Although the idealized calculations here arc
useful to establish the concepts, the exact cal-
culations require running three-dimensional
models on the lull life cycle of convective cloud
systems, followed by validation with detailed
observations.

The importance of the aerosol control of the
released convective energy by adding as much
as 1000 J kg” 1 can be appreciated by consider-
ing that CAPE averages -1000 to 1500 J kg” 1
in the Amazon (30). Simulations of aerosols in-
vigorating peak updrafts by 20% (37, 52) are

www.sciencema 9 .or 9 SCIENCE VOL 321 5 SEPTEMBER 2008

1311

REVIEW

consistent with an increase of released convec-
tive energy by nearly 50%.

Role of Radiative Versus
Microphysical Aerosol Effects

Until now, the radiative and microphysical im-
pacts of aerosols on the climate system have
been considered separately and independently;
their various, often conflicting, influences have
not been amenable to quantitative weighting on
the same scale. Given the opposing microphys-
ical and radiative effects on the vigor and rainfall
amounts of deep warm-base convective clouds,
there is a need to assess the combined effects of
these two factors (25).

A quantitative comparison between the
strengths of the radiative and microphysical ef-
fects of the aerosols is presented in Fig. 4. Be-
cause optically active aerosols are larger than
0.05 pm in radius, and because mature pollu
tion aerosols of this or larger size can act as
CCN (55), CCN concentrations generally in-
crease with aerosol optical thickness (AOT)
(Fig. 1). The empirical relationship between AOT
and CCN is shown in Fig. 4 by AOT - 0.0027 x
(CCHm) 0 -* (1), where CCN 0.4 is the concen-
tration of CCN active at a supersaturation of
0.4%. The cloud droplet concentration N c is
proportional to (CCN 04 )*, where k is typically
smaller than 1 . Using k - 0.825 relates 2000
cloud drops cm -3 to 10 4 CCN 04 cm -3 , which
corresponds to AOT = 1 . The value of k was
inferred from Ramanatlian et al. (7), although
Freud et al. (54) imply that k is closer to 1. In
turn, N c was shown to be related to the depth
above cloud base (D) required for onset of rain
(54). This depth determines the thennodynamic
track of the rising parcel (Fig. 3) and hence the
vigor of the convection and the extent of con-
vective overturning, which detennines the rainfall
amount produced by the cloud system throughout
its life cycle The cloudy parcel ascends along
curve a in Fig. 3 as long as the cloud top has not
reached D, and shifts to a track between curves c
and d according to the amount of condensed
water at that height

The dependence of D on CCN is obtained by a
compilation of aircraft measurements (27, 54, 55)
that provides an approximate relation of D -
80 + (4 x CCN 04 ). According to this relation,
CCN 0.4 should reach -1200 cm -3 for preventing
rainout from typical tropical clouds before reach
ing the practical freezing temperature of -4°C,
which is at D ~ 5 km At this point the in-
vigoration effect is at its maximum, where the
cloud parcel follows curve d in Fig. 3. Adding
CCN beyond this point suppresses the vigor of
the convection by shifting the cloud parcel grad-
ually from curve d to curve a in Fig. 3. This means
that the microphysical effect on invigorating the
convection has a maximum at moderate CCN con-
centrations. This maximum becomes smaller for
cooler-base clouds, where the distance to the freez-
ing level is shorter, so that fewer CCN arc suffi-
cient to suppress the onset of rain up to that level.

At the point of strongest microphysical in-
vigoration, AOT is still at the modest value of
-0.25. Added aerosols increase the AOT and
reduce the flux of solar energy to the surface,
which energizes convection. As a result, with
increasing aerosol loads beyond the optimum,
the weakening of the microphysical invigora-
tion is reinforced by the suppressive effect of
reduced surface heating.

The interplay between the microphysical and
radiative effects of the aerosols may explain the
observations of Bell et al. (55), who showed
that the weekly cycle of air pollution aerosols
in the southeastern United States is associated
with a weekday maximum and weekend min-
imum in the intensity of afternoon convective
rainfall during summer. This was mirrored by
a minimum in the midweek rainfall over the
adjacent sea areas, reflecting an aerosol-induced
modulation of the monsoonal convergence of
air and its rising over land with return flow
aloft to the ocean. This is a remarkable find-
ing, as it suggests that the microphysical im-
pacts of aerosols on invigorating warm-base
deep clouds are not necessarily at the expense
of other clouds in the same region, but can
lead to changes in regional circulation that lead

to greater moisture convergence and regional
precipitation

This weekly cycle emerged in the late 1980s
and strengthened through the 1990s, along with
the contemporary reversal of the dimming trend
of solar radiation reaching the surface, which
took place until the 1980s (56). This was likely
caused by the reversal in the emissions trends
of sulfates and black carbon (57). It is possible
that the weekly cycle emerged when the over-
all aerosol levels decreased to the range where
the microphysical impacts are dominant, as shown
in Fig. 4.

Measuring Radiative and Microphysical
Aerosol Effects with the Same Metric

The precipitation and the radiative effects of
the aerosols (both direct and cloud-mediated)
can be integrally measured when considering
the combined changes in the energy of the at-
mosphere and the surface. The commonly used
metrics arc the radiative forcing at the top of
the atmosphere (TOA) and at the BOA (bottom
of the atmosphere, i.e.. Earth’s surface), measured
in W m -2 . The atmospheric radiative forcing
is the difference between TOA and BOA forc-
ing (7). Here we propose a new metric, the aero-
sol thennodynamic forcing (TF)
(58), representing the aerosol -
induced change in the atmo-
spheric energy budget that is
not radiative in nature. In con-
trast to TOA radiative forcing,
TF does not change the net Earth
energy budget, but rather redis-
tributes it internally; hence, TF
can affect temperature gradi-
ents and atmospheric circula-
tion. The main source of TF is
the change in the amount of latent
heat released by aerosol-induced
changes in clouds and precipi-
tation. It can be expressed as a
change in latent heat flux (in
units of W m~ 2 ) in the atmo-
spheric column.

The vertical distribution of
the atmospheric heating is crit-
ically important because it de-
termines the vertical lapse rate
and hence the CAPE, which
quantifies the ability to produce
convective clouds and precipita-
tion. Atmospheric radiative heat-
ing due to absorbing aerosols
tends to reduce CAPE and there-
by suppress the development
of convective clouds, whereas the
microphysical effects of aero-
sols allow a deeper exploitation
of CAPE and hence invigora-
tion of convection and associated
precipitation.

All the components of the
aerosol radiative (direct and cloud-

■ Transmission
- AOT

3 5
£ S

IT

CCN 04 aerosol concentration (cm 3 )

Fig. 4. Illustration of the relations between the aerosol micro-
physical and radiative effects. The aerosol optical thickness
(AOT) is assumed to reach 1 at CCN 04 = 10 4 cm -3 (dashed red
line), which corresponds to nucleation of 2000 cloud drops cm -3 .
The related transmission of radiation reaching the surface is shown
by the solid red line. The vigor of the convection is shown by
the blue line, which provides the released convective available
potential energy (CAPE) of a cloud parcel that ascends to the
cloud top near the tropopause. The calculation is based on the
scheme in Fig. 3, with respect to curve c as the zero reference.
Note that a maximum in CAPE occurs at CCN 0 .4 = 1200 cm -3 ,
which corresponds to the maximum cloud invigoration accord-
ing to curve d of the scheme in Fig. 3. The AOT corresponding to
the CCN 0.4 at the microphysical optimum is only 0.25. Adding
aerosols beyond this point substantially decreases the vigor of the
cloud because both microphysical and radiative effects work in the
same direction: smaller release of convective energy aloft and less
radiative heating at the surface.

1312

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

REVIEW

mediated) and thermodynamic forcing and the
resulting changes in CAPE can now be quan-
tified as energy flux perturbations in units of
W m“ 2 . Consider the example of smoke chang-
ing tropical convection from thermodynamic
path c to path d in Fig. 3. At the end of the
convective cycle, an additional 1000 J kg~ ; are
depleted from CAPE relative to convection un-
der pristine conditions. The resultant increased
convective overturning is likely to produce more
rainfall and increase the temperature by con-
verting more latent heat into sensible heat, at a
rate of 29 W m” 2 for each added millimeter of
rainfall during 24 hours. This can be consid
ercd as a cloud-mediated TF of aerosols, which
works to enhance rainfall and accelerate the
hydrological cycle, resulting in a positive sign
for TF. On the other hand, if the smoke be-
comes very thick, its radiative impact would
be to reduce surface latent and sensible heat
ing and warm the mid-troposphere. For ex-
ample, an AOT of 1 induces a BOA forcing of
-45 W m 2 in the Amazon (5). This stabiliza-
tion of the atmosphere would cause less con-
vection and depletion of CAPE, less rainfall,
and a resulting deceleration of the hydrolog-
ical cycle (7). Furthermore, too much aerosol
can suppress the precipitation-forming processes
to the extent of changing from thermodynamic
path d to path a in Fig. 3 (see also Fig. 4), hence
reversing the cloud-mediated TF of aerosols from
positive to negative, adding to the negative radia-
tive forcing.

Thermodynamic forcing can occur even with-
out changing the surface rainfall: The energy
change when polluted clouds develop along track
d in Fig. 3, with respect to the pristine reference
state shown in track c, would be defined as TF.
In this case, the TF solely due to added release
of latent heat of freezing is 2.44 W nf 2 mm" 1
day" 1 of heating above the freezing level and
the same amount of cooling due to melting be-
low the melting level. This is a net vertical re-
distribution of latent heat. For an area-average
rainfall of 20 mm day" 1 , the TF scales to 48.8 W
m \ In addition, we should consider the thermo-
dynamic consequences of the aerosol-induced
added rainfall due to increased convective over-
turning. This would convert latent heat to sen-
sible heat at a rate of 29 W m" 2 mm" 1 day" 1 . Such
deeper consumption of CAPE would require a
longer time for the atmosphere to recover for the
next convective cycle, representing a temporal
redistribution of heating and precipitation.

Concluding Thoughts

The next challenge will be to map the radiative
and cloud-mediated thermodynamic forcing of
the aerosols in the parameter space of AOT ver-
sus CCN. The good correlation between AOT
and CCN means that, at least at large scales, die
radiative and microphysical effects of aerosols
on cloud physics are not free to vary indepen-
dently (7), and hence mainly the diagonal of the
parameter space is populated.

According to Fig. 4, there should be an op-
timum aerosol load in the tropical atmosphere
that should lead to the most positive aerosol
thermodynamic forcing, manifested as the most
vigorous convection. This optimum probably oc-
curs at AOT » 0.25 and CCN 0 .4 ~ 1200 cm" 3 .
Remarkably, these fundamental considerations
for AOT ~ 0.25 for optimal cloud development
were matched recently by observations in the
Amazon (59).

This hypothesis reconciles the apparent con-
tradictory reports that were reviewed in two
major assessments ( 18 , 19) as impeding our
overall understanding of cloud-aerosol impacts
on precipitation and the climate system. The
main cause for the previous uncertainties was
the nonmonotonic character of competing ef-
fects, which is inevitable in a system that has an
optimum. The new conceptual model outlined
here improves our understanding and ability
to simulate present and future climates. It also
has implications for intentional weather and
climate modification, which are being consid-
ered in the context of cloud seeding for pre-
cipitation enhancement and geoengineering.
Testing this hypothesis is planned within the
Aerosol Cloud Precipitation Climate (ACPC)
initiative (60, 61).

References and Notes

1 M. 0. Andreae. Atmos. (Trem. Phys Discuss. 8. 11293
(2008).

2. M. 0. Andreae. Science 315. 50 (2007).

3. U. lohmann, J. Feichter, Atmos. Chem. Phys. 5, 715
(2005).

4. V. Ram ana than et at.. Pros NotL Acad. Sti. U.S.A. 102.
5328 (2005).

5. I. Keren. Y. J. Kaufman. L. A Remer. J. V. Martins,
Science 303. 1342 (2004).

6. 0. Rosenfeld, Space So. ftev. 125. 149 (2006).

7. V. Ramanathan. P. J. Crutzen, J. T. Kiehl, D. Rosenfeld.
Science 294. 2119 (2001).

8. S. Menon, J. Hansen. L Nazarenko, Y. F. Luo. Science
297. 2250 (2002).

9. C. Wang. / Geophys. Res. 109. D03106 (2004).

10. Y. Rudich. A Sagi. 0. Rosenfeld. / Geophys Res. 108.

10. 1029/2003 JD00347 2 (2003).

11 R. L. Miller. I. Tegen, J. Pertwitz, / Geophys. Res. 109.
004203 (2004).

12. L D. Rotstayn, U. Lohmann. J. Geophys. Res. 107,
10.1029/2002JD002128 (2002).

13. I. M. Held. T. L Oelworth. J. Lu. K. L Finddl. T. R. Knutson.
Proc. NotL Acad Sd USA 102. 17891 (2005).

14. W. Cotton. R. Pielke, Human Impacts on Weather and
Climate (Cambridge Univ. Press. Cambridge, 2007).

15. R. Gunn. 8. B. Phillips, / MeteoroL 14. 272 (1957).

18. P. Squires, Tetlus 10. 258 (1958).

17. L F. Radke. J. A. Coakley Jr.. M. D. King. Science 248,
1148 (1989).

18. National Research Council. Critical Issues in Weather
Modification Research (National Academies Press.
Washington. DC. 2003).

19. 2. Levin. W. Cotton. Aerosol Pollution Impact on
Precipitation : A Scientific Review. Report from the
WMO/IUGG International Aerosol Precipitation Science
Assessment Group (IAPSAG) (Worid Meteorological
Organization, Geneva. Switzerland. 2007).

20. D. Rosenfeld, Sdence 287. 1793 (2000).

21 0. Rosenfeld. Geophys. Res. left 28. 3105 (1999).

22. D. Rosenfeld. W. L. Woodley, in Cloud Systems.

Hurricanes, and the Tropical Rainfall Measuring Mission
(JRMM), W.-K. Tao. R. Adler. Eds. (American
Meteorological Society, Boston. 2003). pp. 59-80.

23. D. Rosenfeld. R. Lahav. A Khain. M. Pinsky. Sdence 297.
1887 (2002): published online 15 August 2002
(10.1128/science.l073869).

24. Y. Rudkh. 0. Khersonsky. 0. Rosenfeld, Geophys. Res Lett.
29. 10. 1029/2 002GL01605 5 (2002).

25. National Research Council. Radiative Forcing of Climate
Change: Expanding the Concept and Addressing
Uncertainties (National Academies Press. Washington.
DC. 2005).

26. Di Rosenfeld. Y. ]. Kaufman. I. Karen. Atmos Chem. Phys
6. 2503 (2008).

27. 0. Rosenfeld et at.. / Geophys Res. 113. 015203
(2008).

28. M. 0. Andreae et at.. Sdence 303. 1337 (2004).

29. ]. Molinie. C. A. Pontikis. Geophys Res Lett 22. 1085
(1995).

30. E. Williams et at. ). Geophys Res. 107. 10.1029/
2001JD000380 (2002).

31 L Koren. Y. ). Kaufman. D. Rosenfeld. L A Remer.

Y. Rudich. Geophys. Res led 32. L14828 (2005).

32. J. C. Lin. T. Matsui. R. A Pielke Sr, C. Kummerow.

/ Geophys. Res. Ill, D19204 (2006).

33. T. L Bell et al.. / Geophys Res. 113. D02209 (2008).

34. D. M. Schultz, S. Mikkonen. A. Laaksooen. M. B. Richman.
Geophys Res. Lett. 34. L22815 (2007).

35. T. Bell. 0. Rosenfeld. Geophys. Res Lett. 35. 109803
(2008).

36. A Khain. A Pokrovsky. M. Pinsky. A. Seifert. V. Phillips.
/ Atmos. ScL 61. 2963 (2004).

37. A Khain. 0. Rosenfeld. A. Pokrovsky. Q / ff. Meteorol.
Soc 131. 2639 (2005).

38. B. H. Lynn et al.. Mon. Weather Rev. 133. 59 (2005).

39. C. Wang. / Geophys Res. 110. D21211 (2005).

40. S. C. van den Heever. G. G. Canio. W. R. Cotton. P. J. DeMott
A J. Prennl / Atmos. Sd. 63. 1752 (2008).

41 A Seifert, K. 0. Beheng. Meteorol Atmos. Phys. 92. 67
(2008).

42. A Teller. 2. Levin. Atmos. Chem. Phys 6. 67 (2006).

43. W. K. Tao et at.. / Geophys Res. 112, 024S18 (2007).

44. 0. B. Johnson. / Atmos. Sd. 39. 448 (1982).

45. 0. Lohmann. Atmos Chem. Phys. 8. 2115 (2008).

46. L R. Graber. Y. Rudich. Atmos Chem. Phys 6, 729
(2006).

47. V. T. J. Phi Rips, A. Pokrovsky. A. Khain,/ Atmos ScL 64.
338 (2007).

48. M. W. DeMaria. / Atmos Sd. 42. 1944 (1985).

49. 0. Rosenfeld. W. L Woodley. Nature 405. 440 (2000).

50. A P. Khain. D. Rosenfeld. A Pokrovsky. Geophys Res.
Lett 28. 3887 (2001).

51 Z Cui. K. S. Carslaw, Y. Yin. S. Davies. / Geophys. Res
111, 005201 (2006).

52. A P. Khain, N. BenMoshe. A. Pokrovsky. / Atmos. Sri
65. 1721 (2008).

53. U. Dusek et at.. Sdence 312. 1375 (2006).

54. E. Freud. D. Rosenfeld, M. 0. Andreae. A A. Costa.

P. Artaxo, Atmos Chem. Phys. 8. 1661 (2008).

55. M. C. VanZanten. B. Stevens, a Vali, D. H. Lenschow.

/ Atmos. ScL 62. 88 (2005).

56. M. WRd et at.. Sdence 308. 847 (2005).

57. D. G. Streets, Y. Wu, M. Chin. Geophys Res. Lett 33.
L15806 (2006).

58. The term 'thermodynamic aerosol effect* was first
mentioned in (25), but in a more restrictive context.

59. L Koren. J. V. Martins. L A Remer. H. Afargan. Science
321. 946 (2008).

60. B. Stevens. REAPS Newsletter 5. 10 (2008).

61 The Aerosol Cloud Precipitation Climate (ACPC)

initiative is a joint initiative by the International
Geosphere/Biosphete Programme (IGBP) core projects
Integrated Land Ecosystem/Atmosphere Process Study
(iLEAPS) and International Global Atmospheric Chemistry
(IGAO and die World Climate Research Programme
(WCRP) project Global Energy and Water Cycle
Experiment (GEWEX).

62. This paper resulted from discussions held during an ACPC
workshop hosted and supported by the International
Space Science Institute. Bern, Switzerland, through its
International Teams Program.

10.112 6/sdence.ll60606

www.scienceroag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1313

Shadow Enhancers as a Source of
Evolutionary Novelty

]oung-Woo Hong, David A. Hendrix, Michael S. Levine* *

T he dorsal-ventral patterning of the early
Drosophila embryo is controlled by a
sequence-specific transcription tactor, Dorsal,
which is related to mammalian NF-kB (7). Dorsal
works in conceit with two additional transcription
factors. Twist and Snail, to regulate gene expression
in the early embryo. Chromatin immunoprecipitation
(OiIP)-chip assays identified a tew hundred binding
clusters tor Dorsal, Twist, and Snail scattered through-
out the Drosophila genome (2\ Over 35 of these
clusters Junction as authentic enhancers when tested
in transgenic embryos.

ChIPchip assays predicted that many of the
Dorsal target genes contain two separate enhancers tor
the same or similar expression pattern. This prediction
was experimentally confirmed tor \rtd and miR-1 (2).
\iul contains two enhancers that mediate expression in
the presumptive neurogenic ectoderm, whereas miR-1
contains at least two enhancers for expression in
the ventral mesodenn In both cases, the secondary cn
hancers map within 5 kb of tire transcription start site.

However, some of the potential secondary'
enhancers identified by the ChIP-chip assays are

predicted to map quite far from Dorsal target genes.
For example, brinker (brk) is regulated by a known
enhancer located in the 5’ flanking region (3). A
potential secondary enhancer maps within the intron
of a neighboring gene, Alg5, located ~13 kb down-
stream of the brk transcription start site. A ~1 -kb ge-
nomic DNA fragment encompassing the Atg5 intron
was tested tor enhancer activity in transgenic embryos
(Fig. 1A) It directs broad lateral stripes of lacZ re-
porter gene expression, similar to the endogenous
brk expression pattern that is recapitulated by the
previously identified 5' enhancer.

A similar situation is seen tor the Dorsal target
gene, sog. Bioinlbrmatics methods identified an
intronic enhancer that recapitulates the normal sog
expression pattern in the presumptive neurogenic
ectoderm (J). GilP-chip assays identified this en-
hancer, as well as a second cluster of Dorsal, Twist,
and Snail binding sites located 20 kb 5' of the sog
transcription start site, downstream of a neighboring
gene (Fig. IB). The newly identified binding cluster
generates broad lateral stripes of gene expression in
transgenic embryos, similar to those produced by the

intronic enhancer The secondary' enhancers iden-
tified in this study are almost certainly dedicated to
the regulation of brk and sog transcription units be-
cause die associated genes, Alg5 aid CG8J17, re-
spectively , arc not significantly expressed in the early'
embryo (Fig. 1 and fig. SI).

ChIP-chip assays suggest that as many as one-
third or even one-half of all Dorsal target genes might
be regulated by secondary enhancers (2). We propose
the term “shadow enhancer” for remote secondary'
enhancers mapping tar from the target gene and me-
diating activities overlapping the primary enhancer.
Phylogenetic comparisons suggest dial the brk and
sog shadow enhancers are evolving more rapidly
dran the primary enhancers mapping within or near
die two genes (figs. S2 and S3) Despite these dif-
ferent rates of divergence, the overall structures of the
shadow- enhancers are clearly related to their respec-
tive primary enhancers (fig. S4). Given the conser-
vation of die shadow enhancers in all 1 2 sequenced
drosophilids, it is likely that they are essential tor fitness.

Why are Dorsal target genes regulated by shad-
ow enhancers? They might help ensure precise and
reproducible patterns of gene expression during em-
bryogencsis. It is possible that shadow enhancers are
pervasively used in animal development For exam-
ple, the mouse sonic hedgehog gene is regulated in
the floorplate of the embryonic neural tube by two
separate enhancers with slightly’ distinct activities (</).
Shadow enhancers can explain why deletions of well-
defined enhancers sometimes produce no apparent
mutant phenotypes [e.g., (5)). We suggest dial shadow
enhancers might arise from duplication, comparable
to the duplication and divergence of protein-coding
sequences.

The evolution of cis-regulatory DMAs is a major
mechanism of animal diversity [eg., (<5)]. However;
there is the potential problem that such change could
compromise essential genetic activities. Shadow
enhancers have the potential to evolve novel binding
sites and achieve new regulatory activities without
disrupting the core patterning functions of critical
developmental control genes.

References and Notes

L A. Stathopoulov el ai. Cell 111 , 687 (2002).

2. J. Zeitlingc-r el ai. Genes Dev. 21 . 385 (2007).

3. M Maksteirt el at., hoc. Nott. Mod Sci USA 99. 763 (2002).

4. Y. Jeong el at.. Development 133. 761 (2006).

5. N. Xong. C. Kang. D. H. Raulet Immunity 16. 453 (2002).

6. S. Jeong el at.. Cell 132 . 783 (2008).

7. f. Biemai elai. hoc. NatL Mad Sci. US.A 103 . 12763 (2006).

8. This study was funded by the NIH (GM46638) and the

Moore Foundation.

Supporting Online Material

www.sde ncema g. oegfcgi /conte nt/f utl/32 1/58 94/13 14/OC 1

Materials and Methods

Figs. SI to S4

Tables 51 to 53

References and Notes

19 May 2008: accepted 2 July 2008

10.1 126/srience. 1160631

Department of Molecular and Cell Biology, Division of Genetics,
Genomics, and Development, Center for Integrative Genomics,
University of California, Berkeley, CA 94720, USA.

*To whom correspondence should be addressed. E-mail:
mlevine@bedceley.edu

A

B

— " 1

^ ... L F

1 IF

A T ~ If

k k F

j F

„ f

4

— —

yw McZ

yw McZ

Fig. 1. Identification of shadow enhancers. (A) Genome browser showing the Brinker locus (brk) and
neighboring gene AtgS (http^/f lybu 22 .be rkeley.ed tVcg i-bin/g browse/fly4 _3/). The first three lines from the top —
blue, yellow, and red— show the levels of steady-state RNAs in the pipe, ro// mr9/mI °, and ToH 1Cb mutants,
respectively (7). brk transcripts are absent in pipe and Toll ltb mutants but present in ToU m ' 9/rml ° mutants. AtgS is
inactive in all mutants, suggesting that the intronic enhancer is dedicated to brk regulation. The last three lines
show the distributions of Snail, Twist and Dorsal (DD based on whole-genome ChIP-chip assays (2). The leftmost
cluster (open arrowhead) coincides with the known, primary enhancer in the 5' flanking region. A second cluster
(solid arrowhead) is detected 13 kb downstream of the brk transcription start site within Atg5. An ~l-kb genomic
DNA fragment encompassing the 3' binding cluster (shadow enhancer) was tested in transgenic embryos (solid
arrowhead). The 5’ enhancer was tested previously (3). Both fragments function as authentic enhancers to
generate lateral stripes of gene expression in the neurogenic ectoderm. (B) Same as (A) except that the sog locus
is shown, sog transcripts are predominantly detected in ToU rm9/rm10 mutants where the gene is fully active (see the
blue, yellow, and red lines, which show the results of the whole-genome tiling arrays). ChIPchip assays identify
two clusters of Dorsal, Twist, and Snail binding sites within intron 1 and 3' of the neighboring gene, CG8117. Both
genomic DNA fragments function as authentic enhancers to direct lateral stripes of gene expression. Gene
prediction models are displayed above each graphical presentation. Each 3' end is indicated by a triangle.

1314

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

wmmasm

The Crystal Structure of a
Mammalian Fatty Acid Synthase

Mammalian fatty acid synthase is a large multienzyme that catalyzes all steps of fatty acid
synthesis. We have determined its crystal structure at 3.2 angstrom resolution covering five
catalytic domains, whereas the flexibly tethered terminal acyl carrier protein and thioesterase
domains remain unresolved. The structure reveals a complex architecture of alternating linkers and
enzymatic domains. Substrate shuttling is facilitated by flexible tethering of the acyl carrier protein
domain and by the limited contact between the condensing and modifying portions of the
multienzyme, which are mainly connected by linkers rather than direct interaction. The structure
identifies two additional nonenzymatic domains: (i) a pseudo-ketoreductase and (ii) a peripheral
pseudo-methyltransferase that is probably a remnant of an ancestral methyltransferase domain
maintained in some related polyketide synthases. The structural comparison of mammalian fatty
acid synthase with modular polyketide synthases shows how their segmental construction allows
the variation of domain composition to achieve diverse product synthesis.

Timm Maier, Marc Leibundgut, Nenad Ban*

F atty acids fulfill a variety of vital functions:
They are central constituents of biological
membranes, serve as energy storage com-
pounds, and act as second messengers or as co-
valent modifiers governing the localization of
proteins. In bacteria and plants, fatty acid bio-
synthesis is accomplished by a series of mono-
functional proteins in a dissociated type II fatty
acid synthase (FAS) system (7). In contrast, the
type I FASs of fungi and animals arc huge mu!
tifunctional polypeptides that integrate all steps
of fatty acid synthesis into large rnacromolccular
assemblies. Fungal FAS is a 2.6-MD a^p 6 -
heterododecamer with the catalytic domains dis-
tributed over two polypeptides (2-4), whereas
mammalian FAS (mFAS) consists of a 270-kD
polypeptide chain (comprising all seven required
domains) that assembles into homodimers for
enzymatic activity (5, 6).

Despite this variation in structural organiza-
tion, all organisms employ a conserved set of
chemical reactions for fatty acid biosynthesis
(1, 6-8). Stepwise elongation of precursors is
achieved by cyclic decarboxylative condensation
of acyl-coenzyme A (CoA) with the elongation
substrate malonyl-CoA, initiated by the starter
substrate acetyl-CoA. In the priming step, the
acetyl transferase loads acetyl-CoA onto the
terminal thiol of the phosphopantheteine cofactor
of the acyl carrier protein (ACP), which passes
the acetyl moiety over to the active site cysteine
of the p-ketoacyl synthase (KS). Malonyl trans-
ferase (MT) transfers the malonyl group of
malonyl-CoA to ACP, and the KS catalyzes the
decarboxylative condensation of the acetyl and
malonylmoieties to an ACP-bound p-ketoacyl

Institute of Molecular Biology and Biophysics, ETH Zurich,
8092 Zurich, Switzerland.

•To whom correspondence should be addressed. E-mail:
ban@molbiol.ethz.ch

intermediate. The p- carbon position is then mod-
ified by sequential action of the NADPH (the
reduced form of nicotinamide adenine dinuclcotide,
NADP'Vdependent p ketoreductase (KR), a
dehydratase (DH), and the NADPH-depcndcnt
enoyl reductase (ER) to yield a saturated acyl
product elongated by two carbon units. This acyl
group functions as a starter substrate for the next
round of elongation, until the growing fatty acid
chain reaches a length of 16 to 18 carbon atoms
and is released from ACP. In mFAS, the malonyl
and acetyl transferase reactions are catalyzed by a
single bifunctional protein domain, the malonyl-
acetyl transferase (MAT), and the products arc
released from ACP as free fatty acids by a
thioesterase (TE) domain (6).

Humans eating a typical Western diet take in a
surplus of fatty acids. Consequently, de novo
fatty acid biosynthesis and FAS activity arc low
in most body tissues. However, FAS is overex-
prcssed in many cancer cells, and its expression
level is correlated with tumor malignancy (9).
FAS inhibitors have demonstrated anli-tumor
activity in vivo and in vitro, and in recent years
FAS has emerged as an important drug target for
the treatment of human cancer (JO, II). The
medical use of FAS inhibitors has been hampered
by off-target activities. Recently, more specific
inhibitors of type I FAS have been described (12)
and remain to be tested.

Currently, high-resolution structures are
known for all components of bacterial (I) and
fungal FAS (2, 4, 13), whereas the structural in-
formation for mFAS is limited to high-resolution
structures for the isolated MAT [Protein Data
Bank (PDB) entry 2jfd], ACP (14, 15), and TE
domains (16, 1 7) and a domain architecture model
based on a 4.5 A resolution x-ray crystallograph-
ic map (5). Structure determination of KS-acyl
transferase didomain fragments and KR domains
of polyketide synthases (PKS) (18-21 ) — large

modular megasynthases involved in the microbi
al synthesis of a number of bioactive compounds
and drags — has confirmed the anticipated close
structural relation between mFAS and PKS mod-
ules (6, 22). Here, we present the crystal structure
of mFAS in its free and NADP + -bound states, in
which the flexibly tethered C-terminal ACP/TE
domains (23) remain unresolved.

Overall Structure and Topology

The crystal structures of natively purified
mFAS from pigs, free and in complex with the
cofactor NADP + , have been determined at 3.2
and 3.3 A resolution and refined to RJR^ values
of 0.22/0.26 and 0. 19/0.24, respectively (where
R/R^ are El F^h) - F^h) |EF obs (h) cal-
culated for the working/test set of reflections).
Diffraction data were affected by anisotropy with
one weaker direction of reciprocal space (24).
mFAS assembles into an intertwined dimer ap-
proximating an “X” shape (Fig. 1A). This struc-
ture agrees well with our previous architectural
model at intermediate resolution (5) and addi-
tionally provides the connectivities of domains,
the detailed features of active sites, and the nature
of linking sequences outside the conserved core
domains. mFAS is segregated into a lower con-
densing portion, containing the condensing KS
and the MAT domains, and an upper portion
including the DH, ER, and KR domains respon-
sible for P-carbon modification (Fig. 1 , A and B).
Two additional nonenzymatic domains are
located at the periphery of the modifying part.
The first of these domains is homologous to the
methyltransferase family and is thus named
“pseudo-methyltransferase” 0FME). The second
represents a truncated KR fold dimerizing with
the catalytic KR domain and is referred to as
“pseudo-ketoreductase” 0FKR) The condensing
and modifying parts of mFAS are loosely con-
nected and form only tangential contacts. The
structural organization of domains deviates dra-
matically from their linear arrangement in se-
quence (Fig. 1 , A and C).

The two polypeptides dimerize through an
extended contact area of 5400 A 2 , which in-
volves more than 150 residues per chain (table
S2). The main contributions to this interface arise
from homophilic interactions of the KS and ER
domains, with areas of -2600 and 1600 A 2 ,
which resemble the dimer organization of mono-
functional homologs (25, 26). Additional dimer
contacts (800 A 2 ) arc provided by the DH do-
main through homophilic interactions between
the double “hot dog” folds via a loop around
residue 941. The remaining interactions arc
formed by the C-terminal part of the linker re-
gion between the MAT and DH domains (resi-
dues 846 to 860) with the KS domain of the other
chain (400 A 2 ).

Interdomain Linking and Interaction

The characteristics of multienzyme complexes
are to a great degree determined by the nature of

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1315

RESEARCH ARTICLES

the interactions and linking of functional sub-
units. Notably, animal FAS invests only ~9% of
its total sequence tor linkers (Fig. 2, A to E) and
an additional 16% for the lateral noncatalytic
V FME and YKR domains (Fig. 1A). No scat'
folding insertions are found in the catalytic cores.
This is in contrast to the tungal FAS, the other
mcgasynlhase for which a high-resolution struc-
ture is available (2, 4 , 13 ). In that multicnzyme,
almost 50% of the total sequence forms a com-
plex structural matrix of numerous inter- and
intradomain insertions, which define the spatial
organization of the catalytic domains. The only
structured linker domain in mFAS connects the
KS and MAT domains [KS MAT linker domain
(Fig. 2A)] and is composed of amino acids 420 to
490 betw een KS and MAT and of residues S09 to
837 joining MAT and DH. It includes two short
a-helices facing the KS and a three stranded anti-
parallel p sheet on the MAT side and acts as an
adapter, preventing any direct interaction between
the KS and MAT domains. Similar linker domains
were recently found in the KS-acyl transferase
didomain structures of two PKS modules ( 18 , 21 ).
Although an additional helix is inserted in the PKS
linker domains (fig. SI), the relative positions of
the transferase and KS domains remain essentially
the same in mFAS and PKS (fig. S2).

The connection between the condensing and
modifying part of mFAS is provided by residues
838 to 858 between the KS-MAT linker domain
and the DH domain (Fig. 2 , D and E). Again, the
confonnation and position of this linker closely
resembles those observed in KS-acyl transferase
didomain structures from PKS modules, even
though these didomains arc derived from mod-
ules with a considerably different sequence con-
text, containing only KR domains for p-carbon
modification ( 18 , 21 ). Besides the linker itself,
only very limited contacts are formed between
the condensing and modifying parts of mFAS
(with an interaction area of 230 A 2 ). It is even
possible that some percentage of molecules in the
crystal have an alternative connectivity between
the two parts (equivalent to a rotation of the upper
portion of the molecule), which would escape
detection by crystallographic methods (figs. S3
and S4).

The KR domain acts as a central connector
for the modifying part of mFAS and interacts
with the DH, ER, and noncatalytic 'FME and
V KKR domains (Fig. 3, A and B, and table S3). In
contrast, neither the DH nor the ER domain
interacts with either of the noncatalytic domains,
and the contact between the DH and ER domains
is very weak. The KR domain interacts with the

second hot dog subdomain of DH, framing an
800 A 2 interface. The contact between KR and
ER is less intricate and extends over an area of
400 A 2 . About 10% (or 1100 A 2 ) of the KR
surface is involved in a contact with the V FKR
domain, mimicking one of the two major
dimerization interfaces observed in the tetrameric
KR of bacteria (fig. S5). The 4 / ME. which has
the highest mobility based on atomic displace-
ment parameter analysis (fig. S6), protrudes from
the mostly planar body of mFAS. It is docked via
interactions with the KR and the V FKR domain,
the former providing 20% (200 A 2 ) and the latter
80% (800 A 2 ) of the docking area.

Most of the linker regions in the modifying
domains are solvent exposed (Fig. 2). An impor-
tant exception is p strand -forming regions at the
N tenninus of the HTCR-ER linker (residues 1513
to 1518) and the C tenninus of the DH2-'FME
linker (residues 1 1 1 7 to 1 123). These are buried
between the KR, 'FKR, and 'PME domains and
have an important structural role, as discussed in
the next paragraph (Fig. 2 C).

The Nonenzymatic Domains

The KR character of the mFAS V FKR domain,
which has approximately half the size of the
active KR domain, is maintained only in the

Front View

Linear organization

Fig. 1 . Structural overview. (A) Cartoon representation of mFAS, colored
by domains as indicated. Linkers and (inker domains are depicted in
gray. Bound NADP* cofactors and the attachment sites for the disordered
C-terminal ACP/TE domains are shown as blue and black spheres, re-
spectively. The position of the pseudo-twofold dimer axis is depicted by
an arrow; domains of the second chain are indicated by an appended

prime. The lower panel (front view) shows a corresponding schematic
diagram. (B) Top (upper panel) and bottom (lower panel) views,
demonstrating the "S" shape of the modifying (upper) and condensing
(lower) parts of mFAS. The pseudo-twofold axis is indicated by an
ellipsoid. (C) Linear sequence organization of mFAS, at approximate
sequence scale.

1316

5 SEPTEMBER 2008 VOL 321

SCIENCE www.sciencemag.org

RESEARCH ARTICLES

conserved dimerization interface. Because of
extensive truncation of its core, it lias lost the
ability to bind NADPH. Consequently, the V FKR
domain functions mainly to support the integ-
rity of the active site of the catalytic KR domain.
The KR/HTCR arrangement closely resembles
the structure of a KR-'PKR domain derived from
6-deoxyerythronolide B synthase (DEBS) PKS
module 1 (19) (Fig. 3A). The mFAS 'FKR do-
tnain lacks the two N-terminal sheet-helix wind
ings of the DEBS1 'FKR, which itself is already
shorter than the catalytically active KR fold (Fig.
3A). Because of the insertion of the ER and
'FME domains into tire KR YKR fold, two (5
strands originally formed by the linkers flanking
the KR and V FKR domains are no longer directly
adjacent to these two domains in the mFAS
sequence. Rather, they arc provided by amino
acid stretches leading from the DH2 into the
'FME domain, 300 amino acids upstream of the
'FKR, and the linker between 'FKR and ER,
separated by 360 residues from the KR domain
(Fig. 3C).

The 'FME domain is structurally closely re-
lated to S-adenosyl-methionine (SAM)-dependent
methyltransferases, in spite of low sequence ho-

mology (fig. S7A and table S4). The core of
these enzymes consists of a seven-stranded (J
sheet w'ith three helices on each side and the
C-tcrminal strand in anti-parallel orientation (27).
The methyltransferase fold of mFAS carries an
additional short p strand and three helices at its
N terminus (residues 1 125 to 1224). At its C ter-
minus, a short linker (residues 1407 to 1413) leads
directly into the adjacent YKR fold. Such a topol
ogy is characteristic of small-molecule (including
lipid) methyltransferases (27). Nevertheless, the
D/ExGxGxG motif involved in SAM cofactor
binding (27) is not conserved in any of the meta-
zoan FAS sequenced so far (fig. S8). In FAS of
mammals, this motif is changed to ExLxGxG,
which probably prevents co factor binding, in agree-
ment with the absence of methyltransterase ac-
tivity and methylated products in mFAS systems.
Notably, this motif is strictly conserved in several
iterative and modular PKSs found in fungi and
bacteria that share a related overall domain orga-
nization with FAS but are able to methylate
their polyketide substrate with an intrinsic C-
methyltransferase activity (fig. S7B) (28-30).
Thus, the 'FME domain of FAS most likely
represents an inactive version of a previously

Fig. 2. Interdomain
linkers. (A) Surface rep-
resentation of individual
mFAS domains (front
view), colored as in Fig.

L Linking regions are
shown as tubes. (B to E)
Close-up views of indi-
vidual linkers. The direc-
tion of view is indicated
by arrowheads in (A). (B)
Linker connecting the
two subdomains of the
DH domain only loosely
interacts with the main
body of the double hot
dog fold. (0 Linkers in
the KR/ER region are
wrapped around the
domains with close inter-
actions to the domain
surfaces and pronounced
linker-linker contacts;
they mediate interac-
tions between the KR,
WR, and 'FME do-
mains. (0) Modifying
upper and condensing
lower parts of FAS are
only in tangential con-
tact in the region of the
central connection. Few
residues besides the con-
necting linkers mediate
the sparse interactions
via a small interface area.

(G MAT-DH linker mean-
ders through a groove on
the surface of the KS domain.

functional enzyme in a common precursor of
mFAS and PKSs.

Catalytic Domains and Cofactor Binding

Keloacyl synthase. The KS enzymes of all
systems for fatty acid or modular polyketide
synthesis share a common fold and chemical
mechanism, but their substrate specificities differ
considerably (1, 6). In the bacterial systems,
which lack an acetyltransferase, KASHI (FabH)
directly accepts acetyl-CoA as starter substrate.
Further acyl chain extensions from C4 to Cl 4
and from C14 to C16 arc carried out by KASI
(FabB) and KASI I (FabF), respectively. Modular
PKS contain specialized KS with some specific
ity for the p-carbon status but accept a wide range
of substrate lengths (6). In the lungal type I FAS,
only a single KS (which accepts C2 to C16
primers) is required for fatty acid synthesis.
Likewise, mFAS has a single KS domain for all
steps of fatty acid elongation. In contrast to PKS
KS, mFAS KS is highly specific for saturated
acyl chains and does not accept pketoacyl, p
enoyl- or p-hydroxyacyl substrates (6, 31). On
the basis of structural alignments, mFAS KS is
closely related to KS domains from the DEBS I
PKS system [1.3 A root mean square deviation
(RMSD)] but is structurally also very similar to
fungal KS (RMSD 1.8 A). It is more closely
related to bacterial KASI and KASII ( 1 .8 A/1 .6 A
RMSD) than to KASHI (2.9 A RMSD), reflect
ing its ability to elongate long ACP-bound acyl
chains.

Despite the pronounced structural similarity
between the KS domains of mFAS and PKS, the
selectivity of FAS for saturated acyl chains can be
explained by a considerable constriction at the
base of the active site phosphopantetheine bind-
ing pocket leading into the large acyl chain sub-
strate binding tunnel, which connects both active
sites of the KS dimer (Fig. 4, A and B). A number
of residues lining this narrow tunnel are highly
conserved in mFAS but are substituted with
smaller residues in all DEBS PKS modules (Fig.
4B), resulting in a wider, more permissive tunnel
in the KS of DEBS.

Malonyl-acetyl transferase. The acyl trans-
ferases of FAS and PKS systems are composed
of an a/p-hydrolase core fold and a ferredoxin-
like subdomain, which together create the active
site cleft (32). In mFAS, the MAT domains of the
two monomers have slightly different relative
orientations of the two subdomains, probably
selected by crystal -packing interactions. The
structurally closest relatives are acyl transferases
from DEBS didomain structures (2.0 A/2.2 A
RMSD), the human mitochondrial MT (2.2 A),
and tiie Escherichia coli FAS MT (FabD) (2.3 A
RMSD) (table S4). All members of this family
share a conserved active site with a catalytic Ser-
His dyad. MTs are distinguished from acetyl
transferases by the presence of a conserved active
site arginine (2), which forms a bidentate salt
bridge with the malonyl carboxylate (33). The
mFAS MAT and related MTs display a rather

www.sciencemag.org

SCIENCE VOL 321 5 SEPTEMBER 2008

1317

RESEARCH ARTICLES

broad specificity for malonyl-CoA derivatives
(e.g., propionyl-CoA and methyimalonyl-CoA).
However, only mFAS MAT uses both acetyl-
CoA and malonyl-CoA with equal efficiency (6).
A structural comparison with bacterial and PKS
MTs reveals three candidate residues for this
specificity (Fig. 4 C). Phe 682 replaces a serine in
most PKS and bacterial homologs, whereas
Phe 553 and Met -199 substitute for conserved
glutamine residues. Together, these three sub-
stitutions create a considerably more hydropho-
bic active site. The two phenylalanines form a
hydrophobic cavity, which may allow Met' 199 to
flip onto the methyl group of an acetyl substrate.
Thus, the dual specificity of mFAS MAT appears
to result from the combined presence of the
conserved arginine for salt-bridging malonyl
substrates and the more hydrophobic nature of
the active site. The double specificity can be
changed by mutating the arginine to alanine.

which then transforms the MAT into an acetyl
transferase (34).

Ketoreductase. The NADPH-dependent KR
domain belongs to the family of short-chain
dehydrogenases/reductases (SDRs) (55) — single-
domain proteins that have a characteristic
Rossmann fold and a substrate binding extension
inserted before the last helix. mFAS KR is struc-
turally closely related to both the tctrameric
bacterial KR (FabG, 2.0 A RMSD) and ER
(Fabl, 2.2 A RMSD) and to the fungal KR
domain (2.6 A RMSD) (table S4). As for the KS
and MAT domains, the closest structural homol-
ogy is observed with tire KR domains from
modular PKS (RMSDs of 1.5 A for tylosin PKS
KR and 1 .6 A for DEBS KR). The arrangement
of residues in the active site of the mFAS KR
domain is consistent with a proton-relay mecha-
nism described for bacterial FabG (56). However,
two residues of the proton -wire, Asn 2038 and

Lys 1995 , have swapped positions (Fig. 4D), as
previously observed in PKS KR domains (19, 20).
Loops in the vicinity of the active site cleft are
disordered in the apo form of mFAS and become
stabilized upon co factor binding. This includes
residues 1975 to 1990, corresponding to the
|i4/a4 loop in FabG, that are presumably stabi-
lized by interactions of Met 1923 with the active
site Lys 1995 , and part of the substrate binding
extension (residues 2072 to 2075). The direction
of substrate entry into the active site can be in-
ferred from the stereospecificity of mFAS KR,
which produces an ^ hydroxyl group (6, 20). The
substrate approaches the NADPH cofactor from
above the nicotinamide ribosc, as observed for
the structurally related mycobacterial ER InhA
(37,38).

Dehydratase. The mFAS DH domain adopts
a pseudodimeric double hot dog fold (Fig. 5A).
The subdomain arrangement is more similar to

Fig. 3. Modularity of the modifying part of mFAS. (A) Comparison of the
KRAPKR arrangement in mFAS (at left) and a related polyketide synthase,
DEBS1 ( 19 ) (at right). The lower panels provide a schematic overview. The
only modifying domain present in DEBS1 PKS includes KR-TKR, which is
N- and C-terminally extended compared with mFAS KR/'PKR. The
additional ER and DH domains are integrated into mFAS without disturbing
the path of the KRAFKR linker compared with PKS, leaving the N and C

termini in identical positions. (B) KR interacts with all other domains in the
modifying part of mFAS, whereas no direct interactions occur between
either of the nonenzymatic domains and DH or ER. (C) Schematic sequence
diagram depicting the integration and removal of additional domains in
the modifying parts of PKS, mFAS, and insect FAS, on the basis of structural
and sequence alignments (at left). The putative domain topologies of DEBS
modules 1 and 4 are shown schematically on the right

1318

5 SEPTEMBER 2008 VOL 321

SCIENCE www.sciencemag.org

RESEARCH ARTICLES

the functional DH dimers of bacterial FabA/Z
than to the pseudodimeric fungal DH (fig. S9 and
table S4) (39, 40). Each mFAS pseudodimer
contains a single composite active site formed by
residues His 878 from the N-tenninal hot dog fold
and Asp 1033 and His 1037 from the C -terminal fold
(Fig. 5B). The catalytic importance of these
amino acids in mFAS has been demonstrated by
mutagenesis and is further corroborated by a
topologically similar arrangement in the active
centers of bacterial DHs (Fig. 5B) (39-J2). This
also suggests a similar two -base reaction mech-
anism, as proposed for R coli FabA, with His 878
and Asp 1033 participating in substrate protonation
and deprotonation (39). The histidine at position
1037 is only present in the active site of chicken
and pig FAS, whereas in all other sequenced
mFAS the corresponding amino acid is a gluta-
mine (fig. S10). The equivalence of a histidine
and a glutamine at this position has been veri-
fied by mutagenesis (42). In the mFAS structure.
His 1037 is positioned toward Asp 1033 at hydrogen-

bonding distance, indicating a stabilizing func-
tion similar to those of the glutamine in other
mFAS and bacterial DHs (40).

A hydrophobic substrate binding tunnel starts
at the pseudodimer interface, stretches through
the C- terminal hot dog domain, and has an open
end that points toward the top of the FAS
assembly (Fig. 5A). In contrast to type II DHs,
which harbor two equivalent active sites in each
homodimer, the second catalytic site is inactive in
mFAS: The loop harboring the second catalytic
histidine in bacteria is reduced to a short turn, and
the corresponding accessory catalytic residues
located in die central helix of the fold are replaced
by tryptophane and lysine (Fig. 5A). The hy
drophobic tunnel is entirely absent, and the do-
main is truncated by 30 residues at the N
terminus.

Enoylreduclase. In contrast to all other func-
tional domains of the fatty acid elongation cycle,
the mFAS ER has a different fold from its
functional analogs in the bacterial type II FAS

Fig. 4. Active sites of KS, MAT, and KR. (A) A large substrate binding tunnel (blue surface
representation) traverses the dimeric KS domain. The substrate entry site of the KS domains is
oriented toward the MAT domain of the other chain. (B) A narrow constriction in the substrate
binding tunnel of the KS domain adjacent to the conserved active site residues (labeled in black)
prevents the entry of larger modified substrates. Four residues involved in the formation of the
constriction (orange for one subunit and yellow for the second one) are conserved in mFAS and
replaced with smaller residues in the more permissive KS domains of PKS, as exemplified by KS of
DEBS module 5 (purple for one subunit and pink for the second) ( 18 ). (C) MAT active site of mFAS
(red) compared to bacterial MT, with and without bound substrate (light and dark blue,
respectively) ( 32 , 33 ). Conserved active site residues in mFAS are indicated in bold. Two
phenylalanines and a methionine characteristic for the acetyi-CoA/malonyl-CoA double-specific
mFAS create a more hydrophobic binding groove and may close onto the methyl group of an acetyl
moiety to promote efficient binding. (D) Active site of KR with bound NADP* and a 3.3 A unbiased
simulated annealed omit electron density map for the cofactor, contoured at 2.7a. The proton-
donating tyrosine is in equivalent position to bacterial homologs, but the asparagine (N ?038 ) and
lysine (K 1995 ) involved in proton replenishment are swapped.

system, where the ERs are either SDR [FabI,
FahL, FabV (43-45)] or TIM barrel proteins
(FabK) — the latter also found as an ER domain
in fungal type I FAS (2, 46). Instead, the mam-
malian ER establishes a subfamily of medium-
chain dchydrogcnascs/reductases (MDRs) (47)
that is structurally related to bacterial quinone
oxidorcductasc (table S4). The mFAS ER
contains two subdomains, a nucleotide binding
Rossmann-fold (residues 1651 to 1794) and a
substrate binding portion (residues 1530 to 1650
and 1795 to 1858). It binds the NADP + cofactor
in an open extended conformation between the
two subdomains (Fig. 5C). Our structure identi
lies Lys 1771 and Asp 1797 as candidate donor
residues for substrate protonation after hydride
transfer from NADPH (Fig. 5D). These two
residues are in close proximity to the hydride-
donating nicotinamide C4, in a similar position as
the suggested active site tyrosines in other MDR
subfamilies (e.g., the mitochondrial ER) (48-50).
The two residues are strictly conserved in mFAS
(fig. Sll), and a corresponding lysine/aspartate
pair is observed in the apo-form structure of the
nucleotide binding subdomain of a related type I
PKS ER domain (PDB entry lpqw). The active
site of ER is located in a narrow crevice created in
part by the bound nucleotide co factor, very
different from the substrate binding groove in
the related quinone reductase (50). Substrate
entry probably occurs through a tunnel along
the cofactor toward the nicotinamide ring. The
tunnel continues through a constriction toward
the back of the ER domain, where an opening
would allow exit of long acyl chains (Fig. 5C).

Structural Relation to PKS

The structural information presented here pro-
vides extensive evidence for the evolutionary
relationship between mFAS and bacterial and
fungal PKSs. (i) All catalytic mFAS domains are
most closely related to PKS domains at the se-
quence level. Notably, the domains of inFAS arc
more similar to PKS domains than to the bacterial
FAS counterparts, despite the differences in their
substrate specificities, (ii) The structure of mFAS
demonstrates that the similarity to PKSs extends
to the linkers and the overall architecture — e g.,
those in the KS MAT or the KRM'KR regions—
despite very low sequence conservation, (iii) Final-
ly, an additional piece of evidence is provided by
the existence of the nonfunctional H'ME domain,
which can be considered a remnant of a
catalytically active domain present in a common
evolutionary ancestor of mFAS and PKS that is
still preserved in several PKSs (28-30).

Functionally, modular PKSs differ from FAS
by their non-iterative mode of action, where each
module carries out a single precursor elongation
step equivalent to one round of chain elongation
by mFAS (51). The modules are concatenated
into large polypeptides, several of which may
assemble into production lines with more than 10
modules. This allows the synthesis of a variety of
structurally diverse compounds. Based on the

www.sciencema 9 . 0 rg SCIENCE VOL 321 5 SEPTEMBER 2008

1319

RESEARCH ARTICLES

structural similarity discussed above, mFAS can
be considered as a single PKS module spe-
cialized tor iterative fatty acid synthesis.

Individual PKS modules contain substrate
loading and condensing domains and variations
of domains involved in {J-carbon processing that
control the chemical structure of the produced
polyketide. Overall comparison of the PKS
DEBS module 1 KR TKR, which is the only
(I carbon-modi tying domain in this module, with
the corresponding parts of mFAS reveals the
structural basis for the notable modularity of
modifying domains in megasynthases: Mediated
by short linking sequences emanating from con-
served secondary structure elements, the full ER
domain is inserted between the 'FKR and KR
domain, whereas the 'FME domain is integrated
into the linker leading into the N terminus of the
'FKR domain (Fig. 3, A and C). As a conse
quence of this architectural solution, the inser-
tions do not affect the core folds of the 'FKR and
KR domains. Furthermore, because of a very
flexible mode of interaction between KS and DH
and weak contacts between KR and cither DH or
ER (table S3) of the mFAS, it is relatively easy to
envision the architecture of some representative
PKS modules (Fig. 3C). For example, the mini-
mal PKS module, such as module 1 of DEBS 1,
which includes only the KR of the possible
p-carbon-processing domains, would have this
domain linked to KS with a short 10-amino acid
linker, similar to the linker connecting KS to DH
in mFAS. Other truncated variants of the mFAS
architecture are also detectable at the sequence
level. In the case of the closely related insect
FAS, the 'FME domain lacks the N-terminal
extension (Fig. 3C and fig. S8), whereas there is
no methyltransferase in DEBS module 4 (Fig.
3C) ( 19 , 52 ). Short extensions may substitute for
missing domain interactions, as indicated by a C-
teiminal addition of two helices to the KR-'FKR
of DEBS 1 ( 19 ) (Fig. 3A), which covers the re-
gion of the 'FKR surface occupied by the 'FME
in mFAS.

These results also imply that the iterative
mode of elongation (in which ACP shuttles
substrates within one module) and a noniterative
elongation (where substrates are passed between
modules) can be accomplished with a similar
overall architecture of the molecules. Notably,
compared to the mFAS structure, no supplemen-
tary elements with a potential role in oligo-
merization are observed in the PKS structures
available so far, except for a single helix in the
KS-MAT linker domain (fig. SI) and a small C-
tenninal extension of the KR domain (Fig. 3A).
Apparently, the N- and C-terminal docking
domains ( 53 , 54 ) are sufficient to detennine the
higher-order assembly of PKS modules and
polypeptides.

The architecture and the fold of mFAS and of
the related PKSs are extremely versatile. This is
in contrast to fungal FAS, which forms a barrel-
shaped 2.6 MD a 6 p„ heterododecameric assem-
bly with three lull sets of active sites enclosed in

each of the two reaction chambers ( 2 , 13 ). The
scaffold of the cage-forming fungal FAS appears
less tolerant toward product modifying domain
insertions and excisions because of symmetry
constraints and the tight embedding of catalytic
domains. Consistently, no naturally occurring
fungal type I FAS with an altered domain com-
position has been detected, and all fungal FAS
and must of their homologs produce only saturated
fatty acid products ( 8 ).

Substrate Shuttling by the ACP Domain
The entry sites to the active centers of the mFAS
enzymatic domains are grouped around the two
lateral clefts. In each cleft, the entry sites of MAT,
DH, and ER are oriented toward one face of
mFAS, and those of KS and KR toward the other
face (Fig. 6). The flexibly tethered ACP and
the following TE domains are not visualized in
the structure. However, the structure defines the
anchor point of ACP at residue 2113 in the
center of the upper portion of the lateral clefts of
mFAS. Together with recent structures of closely
related rat and human ACP ( 14 , 15 ), which

define the first ordered ACP residue at positions
2125 to 2127 (porcine FAS numbering), the
flexible KR-ACP linker is composed of 12 to 14
amino acids, corresponding to a maximum length
of ~40 A. With a length of 23 to 26 residues, the
ACP-TE linker is substantially longer and could
span up to an 80 A distance, as deduced from the
corresponding isolated domain structures ( 14 - 1 7 ).
In contrast to the fungal ACP linkers, which have
a high Pro/Ala content that can increase their
stiffness (55), the tethers flanking mFAS ACP
have no unusual amino acid composition (Fig. 6).
Whereas the motion of fungal ACP is constrained
by double-tethering, no second anchor point is
apparent for mFAS ACP, because the subsequent
TE domain is not located at a defined position
relative to the body of mFAS. Still, the TE do-
main may influence the motion of the tethered
ACP either by transiently interacting with o titer
domains or by steric effects.

Confining the path of the ACP may be one
role of the protruding noncatalytic 'FME domain.
This domain is docked to the body of mFAS in
the vicinity of the ACP anchor point (Fig. 1, A

Fig. 5. Active sites of DH and ER. (A) Pseudo-dimeric DH domain only harbors a single active site
and substrate binding tunnel with an open end. Active site residues are shown in ball-and-stick
representation. These residues are not conserved in the corresponding position of the second
subdomain (Trp 893 and Lys 897 ). (B) Close-up view of the DH active site topology (top) and
schematic comparison to bacterial and fungal FAS (bottom). The mFAS active site residues (green)
have their functional groups in similar positions as their bacterial FabZ counterparts ( 40 ) (purple),
despite the exchange of two amino acids and the loss of an unusual non-proline cis peptide bond
at H 878 . (C) Open-ended substrate binding tunnel (blue) of the ER domain of mFAS shown in the
presence of the bound NADP* cofactor (yellow spheres). (D) Two amino acids, Asp 1797 and Lys 1771 ,
are candidate proton donor residues for enoyl reduction based on the positioning of their
functional groups at -4.2 A distance to the C4 of the NADP* nicotinamide ring. A 3.3 A unbiased
simulated annealed omit electron density map for the bound NADP* cofactor, contoured at 3.3o, is
shown.

1320

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

RESEARCH ARTICLES

and B) and narrows the accessible region along
the MAT, KS, DH, and ER substrate entry sites to
a rim that is just slightly wider than the A CP (Fig.
6 and fig. SIC). The discrete ACPs of bacterial
type FAS ( 56-58) sequester fully saturated fatty
acyl chains within their hydrophobic core. In
contrast, tnFAS ACP does not bury the acyl chain
inside its core (14), raising the question whether
other pans of the molecule may have taken over
this function (for instance, by providing hydro-
phobic rims for a sliding motion of ACP-tethered
acyl chains on the surface of mFAS). However,
mapping of conservation or electrostatic potential
on the mFAS surface did not reveal such regions.

The linker length, together with the steric
constraints of the mFAS structure, allows ACP to
reach a full set of active sites in one cleft but no
active sites from the other cleft During the
elongation cycle (Fig. 6), the ACP is first loaded
with substrates at the lateral MAT domain. ACP
then has to deliver the substrates to the KS entry

pocket on the opposite face of mFAS. The
shortest route would lead directly through the
cleft (path 1 in Fig. 6), which is just sufficiently
open to allow the passage of ACP. After
condensation at the KS, the ACP must reach the
KR on the same face as the KS active site (path 2)
before crossing the cleft again to approach die
DH domain (path 3). From here, ACP proceeds
toward the nearby ER (path 4) and finally
delivers the fully saturated substrate to the KS
active center before being reloaded at the MAT
for the next cycle. During this cycle, the ACP
interacts with the MAT and the P-caibon-
processing domains of one chain, but it also in
tcracts with the KS of the second polypeptide
chain in the FAS dimer. Notably, the partially
preserved active site cleft of the catalytically
inactive *FME domain could easily be accessed
by ACP from the reaction chamber, as required in
PKSs that display mcthyltransferasc activity (fig.
S7Q (28, 30).

The requirement for ACP to shuttle back and
forth through the cleft does not appear to be the
most favorable solution for efficient substrate
transfer and catalysis by mFAS. An alternative is
suggested by considering the properties of the
junction between the lower condensing and the
upper modifying part of mFAS: They are joined
only via the MAT-DH linkers, which arc ex-
pected to mediate a flexible junction between the
two halves of -200 kD molecular mass each (59).
Moreover, the pscudosymmctry-relatcd DH/KS
contacts on either side of the joint are not iden-
tical as w'ould be expected for a stable interaction
in solution. Consequently, the flexible connection
of the mFAS halves may allow rotational motion
around the dimer axis or a certain degree of tilt-
ing. Such motion would drag the ACP between
the two faces of FAS and may contribute con-
siderably to productive substrate shuttling. Mu-
tant complementation and cross-linking smdies
have demonstrated that the vast majority of sub-

Fig. 6. Substrate shuttling by the ACP in mFAS. After substrate loading at
the MAT on one side of the reaction chamber, the flexibly linked ACP has to
shuttle the substrates to the other side for condensation at the KS and
reduction at the KR (paths 1 and 2). To reach the DH and ER domain, ACP has
to cross the cleft a second time (paths 3 and 4) before the saturated acyl
chain can be back-transferred to KS to serve as primer for the next
elongation cycle. The flexible linkers of ACP are depicted by dashed lines

(pink). The precise length of the KR-ACP linker is defined by the KR and ACP
domain borders in the current and previously solved structures ( 14 , 16 , 17 )
(bottom right, red and green). A schematic representation of the mFAS ACP
domain based on experimentally determined structures ( 14 , 15 ) was
positioned to the active site clefts by superposition with fungal ACP bound
to the KS domain ( 13 ) and by orienting it according to residual electron
density observed in the active site cleft of MAT (fig. S12).

www.sciencema 9 .org SCIENCE VOL 321 5 SEPTEMBER 2008

1321

RESEARCH ARTICLES

stratcs arc processed in mFAS by individual lull
sets of active sites, according to the path of ACP
described above. However, these studies have
also shown that a minority of substrates can be
shuttled between the two sets of active sites,
either by ACP serving both MAT domains or by
direct interaction of ACP with both KS domains
( 6 , 60 - 62 ). In light of the large 135 A distance
between the ACP anchor point located in one
catalytic cleft and the MAT in the other, the most
plausible explanation for the minor mode-of-
domain interaction is a large-scale rotation of the
upper portion of mFAS, relative to the lower
portion (fig. S4).

The molecular description of active sites in
mFAS should stimulate the development of
improved inhibitors as anticancer drug candi
dates. As demonstrated by structural homology,
this structure is also a good template for the
organization of PKS modules; it agrees with and
extends present theoretical models of PKS
architecture ( 19 , 22 ). Furthermore, the structure
of mFAS paves the way for structure-based ex
penmen ts to answer remaining questions on the
dynamics and substrate shuttling mechanism in
megasynthases.

References and Notes

1. S. W. White. J. Zheng, Y. M Zhang, C. 0. Rock. Annu.
Rev. Biochew. 74, 791 (2005).

2. S. Jenni et at.. Science 316. 254 (2007).

3. S. Jenni, M. teibondgut T. Water. N. Ban. Science 311,
1263 (2006).

4. I. B. lontakin, Y. Xiong. T. A Steitt Celt 129. 319 (2007).

5. T. Wafer. S. Jenni, N. Ban. Science 311. 1258 (2006).

6. S. Smith, S. C. Tsai. Not. Prod. Hep. 24. 1041 (2007).

7. F. Lynen, Eur. ). Biochem. 112, 431 (1980).

8. £ Sthweizer. J. Hofmann, Microbiol. Mot. BioL Rev. 68,
501 (2004).

9. F. P. Kuhajda et at.. Proc. Noli Acad. Sti. USA 91.
6379 (1994).

10. ). A. Mervender. R. Lupu. Nat Rev. Cancer 7. 763 (2007).

11. F. P. Kuhajda. Cancer Res. 66. 5977 (2006).

12. H. Orta et at.. Clin. Cancer Re s. 13. 7139 (2007).

13. M. Leibundgut. S. Jenni. C. Frick. N. Ban. Science 316.
288 (2007).

14. L Ploskon ef at.. ). Biot. Chew. 283, 518 (2008).

15. G Bunkoai et at.. Chew. Biol. 14. 1243 (2007).

16. B. Chakravarty. Z. Gu. S. S. Chirala. S. j. Wakil.

F. A. Qu'ocho, Proc. Noll. Acad. Sci. US A 101. 15567
(2004).

17. C. W. Perrble IV. L C. Johnson, S. J. Kridet. W. T. lowther.
Not Struct Mol. Biot. 14. 704 (2007).

18. Y. Tang. C.-Y. IGm. I. I. Mathews. D. L Cane. C. Khosla.
Proc. Natl. Acad. Sci. USA 103. 11124 (2006).

19. A. T. Keatinge-Clay. R. M. Stroud, Structure 14. 737
(2006).

20. A. T. Keatinge-Clay. Chew. BioL 14. 898 (2007).

21. Y. Tang, A. Y. Chen. C.-Y. iGro. D. £ Cane. C. Khosla.
Chew. Biol. 14. 931 (2007).

22. C. Khosla. Y. Tang. A. Y. Chen. N. A. Schnan, D. £ Cane.
Annu. Rev. Biochem. 76. 195 (2007).

23. A. K. Joshi. A. Witkowski. H. A Berman. L. Zhang. S. Smith.
Biochemistry 44, 4100 (2005).

24. Materials and methods are available as supporting
material on Science Online.

25. J. G. Olsen. A. Kadriola. P. von Wettstein-Knowles.

M. Siggaard Andersen. S. Larsen, Structure 9. 233 (2001).

26. J. M. Thorn. J. D. Barton. N. £ Dixon. D. L OIK
K. J. Edwards, J. Mot. Biot. 249. 785 (1995).

27. J. L Martin. F. M. McMillan. Curt. Opin. Struct Biot. 12.
783 (2002).

28. I. Fujii, N. Yoshida. S. Shimomaki. H. Oikawa, Y. Ebiruka.
Chew. Biol. 12, 1301 (2005).

29. !. Molnar el at.. Chew. Biol. 7. 97 (2000).

30. D. J. Edwards ef at.. Chem Biot. 11. 817 (2004).

31* A. Witkowski, A K. Joshi, S. Smith. Biochemistry 36.

16338 (1997).

32. L Seme. £ C. Verbree, Z. Dauter. A. R. Stuitje.

Z. S. Derewenda . ). BioL Chem. 270. 12961 (1995).

33. C. Oefner, H. Schulz. A. D Arcy. G E. Dale. Acta
Crystatlogr. 0 Biol. CrystoUogr. 62. 613 (2006).

34. V. S. Rangan, S. Smith./. Biot. Chem. 272. 11975(1997).

35. B. Persson, Y. Kallberg, J. Oppermann. H. JomvalL
Chem. Biol. Interact. 143 144, 271 (2003).

36. A. C. Price. Y. M. Zhang, C. 0. Rock. S. W. White,
Stnrcture 12, 417 (2004).

37. 0. A Rorwarsfci. C. Vilcheze. M. Sugantino, R. Bittman.

J. C. Sacchettini. /. BioL Chem. 274. 15582 (1999).

38. T. J. Sullivan ef oL Am. Chem. Soc. Chem Biol. 1. 43
(2006).

39. M. Leesong. B. S. Henderson. ). R. Oilrig. J. M Schwab.

J. L. Smilh. Structure 4. 253 (1996).

40. M. S. Kimber ef at.. }. BioL Chem. 279. 52593 (2004).
41 A. K. Joshi. S. Smith. / BioL Chem. 268. 22508 (1993).

42. S. Pasta. A Witkowski. A It Joshi. S. Smith, Chem Biol.
14. 1377 (2007).

43. C. Baldodc J. B. Rafferty. A R. Stuitje. A. R. Slabas.

D. W. Rice. /. Mol. Biot. 284, 1529 (1998).

44. R. J. Heath. N. Su. C. K. Murphy. C. 0. Rock . ). BioL
Chem. 275. 40128 (2000).

45. R. P. Masse ngo-Tiasse. J. t Cronan, ]. BioL Chem. 283.
1308 (2008).

46. J. Saito ef at.. Protein Sci. 17. 691 (2008).

47. E. Nordling, H. Jornvall. B. Persson. Eur. ). Biochem. 269.
4267 (2002).

48. T. T. Airenne ef at. }. Mol. Biot. 327. 47 (2003).

49. T. Hori ef at.. }. Biot. Chem 279. 22615 (2004).

50. Y. Shimomura, Y. Kakuta. K. Fukuyama, /. Bacterial. 185.
4211 (2003).

51 D. A Hopwood. D. H. Sherman, Annu. Rev. Genet 24. 37
(1990).

52. C. D. Richter et at.. EEBS ). 274. 2196 (2007).

53. C. D. Richter. D. Nietlispach, R. W. Broadhurst,

K. J. Weissman, Nat Chem. Biol. 4. 75 (2008).

prominent theory states that the hippo-
campal system primarily serves spatial
navigation (/, 2); a component of this
theory is that the place-dependent activity of
neurons [place cells ( 1 , 2 )\ in the hippocampus
arises from external serially ordered environ-
mental stimuli ( 3 - 7 ). Place cells arc thought to
embody the representation of a cognitive map,
enabling flexible navigation. However, neural
theories of other cognitive processes that may
depend on the hippocampus, such as episodic

54. K. J. Weissman. ChemBioChem 7. 485 (2006).

55. R. N. Perham. Biochemistty 30. 8501 (1991).

56. A Roujeinikova ef at.. Structure 10. 825 (2002).

57. A Roujeinikova ef at.. ). MoL Biot. 365. 135 (2007).

58. G A. Zometzer, B. G Fox. J. L. Maiktey. Biochemistry 45.
5217 (2006).

59. S. Jones. J. M. Thornton. Proc. Natt Acad. Scl USA 93.
13 (1996).

60. S. Smith. A Witkowski. A K. Joshi. Prog, lipid Res. 42.
289 (2003).

61. A K. Joshi. V. S. Rangan. A Witkowski. S. Smith. Chem.
Biot. 10. 169 (2003).

62. V. S. Rangan. A. K. Joshi. S. Smith. Biochemistry 40.
10792 (2001).

63. AH data were collected at the Swiss Light Source (5LS,
Paul Schemer Institute, Villigen). We thank

C. Schulze Btiese, S. Gutmann. R. Bingel Erlenmeyer.

S. Russo. A Pauluhn. and T. Tomizaki for their
outstanding support at the SLS; S. Jenni and M. Sutter for
critically reading the manuscript and all members of the
Ban laboratory for suggestions and discussions;

R. Grosse-Kunstleve, P. Aionine. and P. Adams for
support with the PHENIX software: and A Jones for
support with the program 0. This work was supported by
the Swiss National Science Foundation (SNSF) and the
National Center of Excellence in Research Structural
Biology program of the SNSF. Structure factors and
atomic coordinates of the porcine FAS in the apo and
NADP* bound form have been deposited in the Protein
Data Bank with accession codes 2vz8 and 2vz9-

Supporting Online Material

www.scie ncema g.or gfcgi/conte nt/fult/32 1/58 94/13 15/DC 1

Materials and Methods

Figs. SI to S15

Tables Si to S4

References

3 June 2008: accepted 31 July 2008
10.112 6/science.ll61269

memory and action planning, draw on the activ-
ity of hypothetical internally organized cell as-
semblies ( 8 - 13 ).

Center for Molecular and Behavioral Neuroscience, Rutgers,
The State University of New Jersey, 197 University Avenue,
Newark, NJ 07102, USA

•Present address: Center for Neurobiology and Behavior,
Columbia University, 1051 Riverside Drive, New York, NY
10032, USA

fTo whom correspondence should be addressed. E-mail:
buzsaki @axon.rutg ers.edu

Internally Generated Cell Assembly
Sequences in the Rat Hippocampus

Eva Pastalkova, Vladimir Itskov,’ Asohan Amarasingham, Gyorgy Buzsakif

A long-standing conjecture in neuroscience is that aspects of cognition depend on the brain's ability
to self-generate sequential neuronal activity. We found that reliably and continually changing cell
assemblies in the rat hippocampus appeared not only during spatial navigation but also in the
absence of changing environmental or body-derived inputs. During the delay period of a memory
task, each moment in time was characterized by the activity of a particular assembly of neurons.
Identical initial conditions triggered a similar assembly sequence, whereas different conditions
gave rise to different sequences, thereby predicting behavioral choices, including errors. Such
sequences were not formed in control (non memory) tasks. We hypothesize that neuronal
representations, evolved for encoding distance in spatial navigation, also support episodic recall
and the planning of action sequences.

1322

5 SEPTEMBER 2008 VOL 321 SCIENCE wvwv.sciencemag.org

RESEARCH ARTICLES

Several observations have refined the navi-
gation theory. Hippocampal neurons can predict
where the animal is coming from, or its desti-
nation ( 14-17)\ the sequential activity of place
cells during locomotion is replicated within
single cycles of the theta oscillation (8 to 12 Hz)
(18-20): furthermore, the temporal recruitment of
active neurons in the population bursts of rest and
sleep also reflects, again on a faster time scale,
their sequential activity as place cells during
locomotion (21-23). Thus, the sequential activa-
tion of hippocampal neurons can be disengaged
from external landmarks (24, 25). However, in-
ternally generated assembly sequences operat-

ing at the time scale of behavior have not yet
been reported.

The frameworks of environment-controlled
versus internally generated assembly sequences
give rise to distinct predictions. Imagine that a rat
is frozen in position during its travel (and yet the
theta oscillation is maintained). According to the
navigation theory, a subset of landmark-controlled
place cells should then display sustained activity,
and other neurons would remain suppressed (2-6).
In contrast, if assembly sequences were gener-
ated by internal mechanisms, neurons might
rather display continually changing activity. VVe
tested these predictions by examining the activity

of hippocampal neurons while a rat was running
in a wheel at a relatively constant speed (26, 27)
during the delay of a hippocampus-dependent
alternation memory task.

Internally generated cell-assembly sequences.
Rats were trained to alternate between the left
and right arms of a figure-eight maze [Fig. 1 A
and supporting online material (SOM) text).
During the delay period between maze runs (10 s
for rat 1; 20 s each for rats 2 and 3), the animals
were trained to run steadily in the same direction
in a wheel (Fig. 1 A). To confront the predictions
of the navigation theory with those of the internal
sequence-generation hypothesis, we compared

Fig. 1. Episode fields in
the wheel and place
fields in the maze are
similar. (A) Cobr-coded
spikes (dots) of simulta-
neously recorded hippo-
campal CA1 pyramidal
neurons. The rat was re-
quired to run in the
wheel facing to the left
during the delay be
tween the runs in the
maze. (B) Percent of
neurons firing >0.2 Hz
within each pixel The
highest percentage of
neurons was active when
rats were running in the
wheel. (O Relationship
between firing rate of
neurons active in rats
running the wheel and
the maze (r s = -0.3, P <
0 . 0001 , 681 neurons,
three rats, 17 sessions).
(D) Normalized firing
rate of six simultane-
ously recorded neurons
during wheel running
(each line shows the
cobr-coded activity on
single trials turning to
the left arm). The epi-
sode fields occurred at
specific segments of the
run. (E) Normalized fir-
ing rate of 30 simulta-
neously recorded neurons
during wheel running,
ordered by the latency
of their peak firing rate.

(F) Width (top) and peak
firing rate (bottom) of
episode and place fields
(wheel n = 135 neurons;
maze, n = 162 neurons).
Arrows indicate medians.

(G) Population vector
cross-correlation matrix

Time in wheel (sec)

G

Time in wheel (sec)

Time (sec)

Time (sec)

(SOM text). The width of the diagonal stripe indicates the rate at which neuronal assemblies transition, (lower left) The decay of the population vector correlation
during wheel running and maze traversal Thin lines, individual sessions; thick lines, group means.

www.sciencema 9 .or 9 SCIENCE VOL 321 5 SEPTEMBER 2008

1323

RESEARCH ARTICLES

the firing patterns of CA1 hippocampal neurons
in rats running the wheel and the maze.

We analyzed the activity of -500 pyramidal
cells recorded in the wheel and -600 neurons in
the maze (mean firing rate >0.5 Hz) (Fig. 1A).
Pyramidal neurons were transiently active in rats
running both tire maze [place cells (7)] and the
wheel. Although the position and direction of the
rat’s head were stationary' during wheel running
(fig. SI), the percentage of neurons active in the
pixels occupied by the head during wheel run-
ning was three to four times greater than in any
area of comparable size in the maze (Wilcoxon
rank sum test, P < 0.0001) (Fig. IB). Thus, if
pyramidal neurons were solely activated by
environmental cues (2-6), this finding would
reflect several- fold-stronger neuronal representa-
tion of the animal’s position within the wheel.
Many individual pyramidal cells were active both
in rats running the wheel and rats running the
maze, but the sequential order of their activation
in rats in the wheel was unrelated to that of rats
in the maze, and their firing rates in these two
areas were inversely correlated [Spearman corre-
lation coefficient (r,) - -0.3, P< 0.0001, n - 681
neurons (Figs. 1C and 4B); contrast this with the
population of intemeurons, r t = 0.85, P< 0.0001,
n - 125 intemeurons (fig. S2)]. The average
proportion of pyramidal neurons simultaneously

active [firing at least a single spike in 100-ms
windows (averaging over 100-ms windows)]
was similar in the wheel (10.75 ± 3.97%) and
the maze (12.56 ± 4.32%) (fig. S3).

Pyramidal neurons typically fired transiently,
and reliably in successive trials, at specific times
of wheel running (episode fields), and most cells
had multiple peaks of varying sizes (Fig. ID).
Typically, and reminiscent of a synfire chain (II),
at least one episode cell was active at every
moment of a wheel run (Fig. IE).

Were episode cells in rats in the wheel
generated by the same mechanism as place cells
in rats in the maze? We looked for evidence of
differing mechanisms by comparing several mea-
sures of the firing of episode and place cells.
First, we calculated the duration of activity (field
width) (Fig. IF) of single cells [including only
fields with a peak firing rate of >6.0 Hz and >4.5
SD above the mean firing rate (SOM text)]. The
temporal and spatial extent of the field was
determined as those times and positions at which
firing rates were at least 10% that of the peak
firing rate (in the wheel or maze) (19, 28). By
these criteria, 32% of the neurons recorded in the
wheel and 22% in the maze had at least one field.
Neither the distribution of field widths (medians
were 0.94 and 1 .0 s, respectively; Wilcoxon test,
P = 0.44) nor peak firing rates (medians were

13.08 and 12.8 Hz, respectively; P = 0.61)
differed significantly between the episode and
place fields (Fig. IF). Second, to measure the
average lifetime of assembly activity for a pop-
ulation, we determined the maximal time lag at
which the autocorrelation of the population’s ac-
tivity was above 0.5 (29) and again found no
significant difference, with respect to the median,
between the populations of episode and place
cells (medians were 0.83 and 0.75 s, respectively;
P = 0.32) (Fig. 1G). Third, we compared the
relationship between spikes and the local field
potential in episode and place cells. On linear
tracks, sequentially generated spikes of a place
cell gradually shift to earlier and earlier phases of
the theta oscillation as the rat passes through the
place field (phase precession), and there is a sys-
tematic relationship between the phase of spikes
and the animal's position (3, 18-20, 28, 30, 31).
The navigation theory predicts that the phase of
spikes will remain fixed if environmental inputs
do not change (3, 26, 27). In contrast, episode
cells displayed phase precession during wheel
running (Fig. 2A). Similarly to place cells, the
theta frequency oscillation of episode cells was
higher than that of the field theta rhythm (Fig.
2B), and the slope of the phase precession was
inversely related to the length of the episode
field (Fig. 2, A and D) (3, 19, 20, 28, 30, 31).

\Aa/Wv\AAA

Fig. 2. Episode neurons
in the wheel display
theta phase precession
and temporal compres-
sion. (A) (Top) Unfiltered
(light gray) and filtered
(4 to 10 Hz) (dark gray)
traces of LFP and phase
advancement of action
potentials (dot). (Bottom)

Activity of six example
neurons from the same
session. Each dot is an
action potential, displayed
as a function of theta
phase and time from the
beginning of wheel run-
ning from all trials. One
and a half theta cycles
are shown (y axis). Red
line, smoothed firing
rate. (B) Power spectra
of spike trains generated
during wheel running
( n = 283 pyramidal neu-
rons) and the simulta-
neously recorded LFP.

Faster oscillation of neu-
rons occurs relative to
LFP. (C) Slope of theta
phase precession within
episode fields in the

wheel and within place fields in the maze. (D) Relationship between phase
precession slope and episode length (left, r t = 0.46, P < 0.0001) and
episode field width (right, r t = 0.52, P < 0.0001), respectively. (E)
Temporal compression of spikes sequences. Correlation of the distance

-0 25 0 0 25

A distance (m)

between the peaks of episode fields of neuron pairs in the wheel with
the temporal offset of the pair's cross-correlogram peaks is shown.
Each dot represents a neuron pair (n - 105 eligible pairs; three rats; r, =
0.59; P < 0.0001).

1324

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

RESEARCH ARTICLES

Furthermore, the slopes correlated more strongly
with tire length of the episode field (r, - 0.52, P<
0.0001) than with the time it took the rat to run
through the same field (r* = 0.46, P < 0.0001)
(Figs. 2D and 3) because of the variability of the
rat’s running speed (28). The distributions of
phase precession slopes for the episode and place
fields were also similar (medians were -0.6°/s
and -0.6°/s, respectively; P = 0.6) (Fig. 2 C).
Finally, we compared the spike timing relation-
ships among neurons. During maze traversals,
the distance between the place-field peaks of a
neuronal pair was correlated with the temporal
offset between its spikes within the theta cycle, a
phenomenon known as distance-time compres-

sion (SOM text) (18, 19). Analogously, the
distance between peaks of the episode fields of
neuron pairs (episode fields with peak firing rate
>5 Hz and >3 SD above the mean firing rate were
included in this analysis; n - 105 pairs) was
correlated with the temporal offsets between the
spikes at the theta time scale (r t - 0.59, P <
0.0001) (Fig. 2E). These findings indicate that
the mechanisms generating place and episode
fields are similar.

Body cues are not sufficient to generate
assembly sequences. It has been suggested that
in addition to generating a cognitive map of the
environment (2), tire hippocampus and its asso-
ciated structures integrate self-motion-induced

information (7, 32, 33). Were the episode cell
sequences generated by idiothetic self-motion
cues? We examined population firing patterns in
two control (nonmemory) tasks. In the first task
(control I), the animals (rats 3 and 4) were
required to run in the wheel for a water reward
available in an adjacent box (26). In tire second
task (control 2), the animals (rats 2 and 3) had
continuous access to a wheel adjacent to their
home cage, and recordings were made during
spontaneous wheel-running episodes. Transient
firing patterns, consistent across trials, were
rarely observed during the control tasks. Rather,
the majority of active neurons exhibited relative-
ly sustained firing throughout tire wheel -running

a i • • ►*

i£ \

2 4 6 8 1U z 4

Time in wheel (sec)

% : ►

r^r

*4/-' ■ ; •

• M T*

* &

— •

^ • .* -*

1 ; *

4

3

jBI

2.5 5 75 If

.. _

> 2.5 5 7.5 K

) 2.5

5 7.5 1

W * 0 ' 6 I

Time in wheel (sec)

Frequency (Hz)

A frequency (un'rt-LFP; Hz)

% spikes within ACG peak

Fig. 3. Filing patterns
during wheel running
depend on the context
of the task. (A) (Top)

Activity of representative
single neurons (color-
coded) during wheel run-
ning in control tasks 1 and
2 (compare with Fig. ID).

(Bottom) Unit discharges
(dots) from all trials with-
in a session as a function
of theta phase, plotted
against time from the be-
ginning of a wheel run.

Red line, smoothed mean
firing rate. Relatively
steady firing rates and a
steady theta phase occur
in both control tasks. (B)

Cross-correlation matrices
in three different tasks
(memoiy and control 2
are from the same rat). In
the memory task, trials
with the same future
choices [left (L>— trials^
versus L -trials,,,, x and
right (R}-trials„ versus
R-trials„*i) were cross-
correlated, whereas in
control tasks trials,, and
trials„„i were cross-
correlated. Only pixel
values significantly dif-
ferent from chance are
shown (Spearman rank
correlation, P < 0.01).

(O Population-vector cor-
relation coefficient values
in the memory task (n =

17 sessions) and control
tasks 0? = 8 sessions)

(mean ± SD). (D) Power spectrum of spike trains of an episode neuron (unit)
and simultaneously recorded LFP during wheel running in the memory task
(30). The frequency of unit firing oscillation is higher than the frequency of
LFP. (E) Difference between unit and LFP oscillation frequency in the
memory (left) and control (right) tasks. Each line is a color-coded normalized
cross -cor relog ram between power spectra of a pyramidal neuron and
simultaneously recorded LFP. A shift of the maximal correlation values to the
right indicates that unit theta oscillation is faster than LFP theta oscillation

(black dots, maxima of the cross-correlograms; white line, sum of all
neurons). There is a significant frequency shift in the memory task (0.44 ±
0.6 Hz) and a lack of frequency shift in control tasks (combined control 1 and
2, 0.07 t 0.3 Hz). (F) Ratio of spikes in the center and tail of temporal auto-
correlograms (SOM text). High values indicate compact episode fields; low
values indicate spikes scattered throughout the time of wheel running
(memory task, n = 287 neurons; control tasks, n = 85 neurons; rank sum test,
P< 0.0001). Arrows indicate medians.

www.sciencema 9 .or 9 SCIENCE VOL 321 5 SEPTEMBER 2008

1325

RESEARCH ARTICLES

Time in wheel (sec)

Fig. 4. Cell-assembly activity in the wheel predicts the future choice of the rat in the maze. (A)
Examples of three neurons that strongly differentiated between wheel-running trials preceding
right and left choices (fig. S7 and movie SI). (B) Normalized firing rate profiles of neurons during
wheel running and in the stem of the maze, ordered by the latency of their peak firing rates during
left trials (each line is a single cell; cells are combined from all sessions). White line, time gap
between the end of wheel running and the initiation of maze stem traversal. (Middle) Normalized
firing rates of the same neurons during right trials. (Right) Time periods of significant differences
(P < 0.05) in firing rates between left and right trials for respective neurons (red line, R > L; blue
line, L > R). Gray line, number of neurons discriminating between left and right trials as a function
of wheel-running time.

periods (Pig. 3A and fig. S4) (5, 26 - 27 ). During
runs of opposite direction in the wheel, different
populations of neurons were active (fig. S5)
( 26 ), arguing for the importance of distant cues
( 2 , 20 ) and against a critical role of idiothetic
inputs ( 26 ). In addition, the temporal organiza-
tion of cell assemblies in control tasks was less
precise, as reflected by much weaker correla-
tions between temporally adjacent populations
during the control tasks than during the memory
task (Fig. 3, B and C), despite the similarity in
firing rates during all tasks (fig. S6). As another
contrast to the memory task, neurons recorded
during tire control tasks fired throughout tire
trial, with spikes locked to a similar phase of the
theta cycle (Fig. 3A). Consistent with these
observations, neurons in the rats perfonning the
memory task oscillated faster than the local held
potential (LFP) [difference (A) = 0.44 ± 0.6 Hz)
(Figs. 2B and 3, D and E), an indication of
phase precession ( 19 , 20 , 29 , 30 ), whereas dur-
ing the control tasks, the power spectra of the
units and LFP were similar (A - 0.07 ± 0.3 Hz)
(Fig. 3E). Finally, to quantify differences in
temporal clustering of spikes, we examined an
autocorrelogram of each neuron. We applied
(after filtering, 0.2 to 2 Hz) the same definition
for the peak region boundaries that we used for
the episode field detection boundary of the epi-
sode field (the 10% boundary) and then com-
pared, for each neuron, the ratio of the number
of spikes that fell within the peak region bound-
ary to those that fell outside. These ratios were
significantly larger during tire memory task and
reflected the temporal compactness of firing
during the memory task as opposed to the con-
trol tasks (Fig. 3F). Thus, the indicators of tem-
porally precise sequential activity in neuronal
populations were absent during the control
tasks, despite indistinguishable motor character-
istics across all tasks.

Assembly sequences depend on memory
load. What is the behavioral function of in-
ternally generated cell-assembly sequences?
Temporarily inactivating neuronal circuits in the
dorsal hippocampus, we found that performance
in the delayed alternation task depends on the
integrity of the hippocampus (fig. S7) ( 17 ). Titus,
we hypothesized that infonnation about choice
behavior is reflected in assembly sequences ( 34 ).
All correctly performed trials were sorted accord-
ing to the rat’s future choice of arm (left or right),
and choice-specific firing effects were identified
by comparing the firing patterns of single neu-
rons with those of surrogate spike trains created
by shuffling the left and right labels (Fig. 4, A
and B, and SOM text) ( 34 ). Some neurons were
active exclusively before either the left or right
choice, whereas others showed differential firing
rates and/or fired at different times after the
beginning of wheel running (Fig. 4A, figs. S8 to
S10, and movie SI). The largest proportion of
neurons exhibiting choice-predictive activity was
at the beginning of the run; this proportion de-
creased as a function of time during the delay and

in the stem of the maze (Fig. 4B), suggesting a
critical role for initial conditions in specifying the
sequences (fig. Sll). In addition, we designed a
probabilistic model of the relationship between
neuronal firing patterns and the animal’s choices
(SOM text). Using this model, the accuracy of
single-trial prediction, under cross-validation,
varied horn low (near 50%) and not significant
to 100% and significant across many sessions
(fig. S9).

Because the rat was perfonning an alternation
task, past and future choices were deterministi-
cally related on correctly perfonned trials, and it

was not possible to disambiguate their influence
on neuronal activity. To distinguish such retro-
spective and prospective factors ( 14 - 17 ), we
examined cell-assembly sequences during error
trials. Neurons that reliably predicted the behav
ioral choice of the rat on correct trials continued
to predict the choice behavior on error trials (Fig.
5 A, tig. SI 2, and movie SI) ( 15 , 24 ). Similarly,
population sequences that differentiated correct
behavioral choices continued to predict behav-
ioral choice errors (Fig. 5, B and C, and fig. SI 3).
Although there were only a few error trials, a
majority of them could be predicted from the

1326

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

RESEARCH ARTICLES

A

S 10 IS S 10 IS

Time in wheel (sec)

Fig. 5. Cell-assembly activity during the wheel predicts behavioral errors
during the maze. (A) Two example neurons from a session with seven left
error trials (err). Correct trials are separated into left- and right-turn trials.
(B) Normalized firing rates of 43 neurons simultaneously recorded during
wheel running, ordered by the latency of peak firing rates during correct
left trials (left). (Right) Firing sequence of the same neurons on cored
right trials. (C) Firing sequence of neurons in a single error (left) trial
Neuronal order is the same as in (B). The firing sequence during the error
trial is similar to that of the correct left trials. The correlation coefficient
between correct and error trial sequences is 0.45 (fig. S13). (D) Percent of
correctly predicted errors from the neuronal population activity.

B Session mean (correct left)

Session mean (correct right)

wheel (sec)

Time in wheel (sec)

firing panems of neurons during wheel running
(Fig. 5D). Altogether, these observations demon-
strate that a particular sequence of neurons was
activated in a reliable temporal order from the
moment the rat entered the wheel to the time it
reached the reward.

Because running speed, head position, and head
direction during wheel running before left and right
choices were apparently indistinguishable (fig. SIX
the above findings indicate that trial differences
in hippocampal assembly configurations cannot
solely arise from instantaneous environmental
inputs or the integration of motion signals.

Behavioral function of internally gener-
ated cell-assembly sequences. These findings
demonstrate that the rat brain can generate con-
tinually changing assembly sequences. The pat-
terns of the self-evolving neuronal assembly
sequences depend on the initial conditions, and
the particular sequences of cell assemblies are
predictive of behavioral outcome.

Our results offer new insights into the rela-
tionship between hippocampal activity and
navigation (2-7, 14 - 20 , 26 - 30 , 33 ). Hippocam-
pal firing patterns during maze navigation were
similar to those during wheel running in the
delayed alternation memory task with stationary
environmental and body cues. Therefore, we
suggest that hippocampal networks can produce
sequential firing patterns in two possibly interact-
ing ways: under the influence of environmental/
idiothetic cues or by self-organized internal mech-
anisms. The high-dimensional and largely ran-
dom (nontopographical) connectivity of the CA3
axonal system ( 35 ) and its inputs makes the
hippocampus an ideal candidate for internal se-
quence generation ( 13 , 33 , 36 , 37 ). The parame-
ters of cell- assembly dynamics (including their
trajectory and lifetimes) are probably affected by
a number of factors, including experience-
dependent and short-term synaptic plasticity
( 34 , 38 )\ asymmetric inhibition (39); brain state;

and, fundamentally, the character and context of
the input. The evolving trajectory can be ef-
fectively perturbed, or updated, by external inputs
in every theta cycle ( 40 ). Because of this
flexibility in the sources of cell-assembly control,
we hypothesize that neuronal algorithms, having
evolved tor the computation of distances, can also
support the episodic recall of events and the
planning of action sequences and goals ( 19 ).
During learning, the temporal order of external
events is instrumental in specifying and securing
the appropriate neuronal representations, whereas
during recall, imagination (35), or action plan-
ning, the sequence identity is determined by the
intrinsic dynamics of the network.

References and Notes

1. J. O'Keefe. J. Dostrovsky. Brain Res. 34. 171 (1971).

2. J. O'Keefe. L Nadel. The Hippocampus os a Cognitive
Mop (Clarendon. Oxford, UK. 1978).

3. J. Hotter. N. Burgess, J. O'Keefe. Nature 425. 828 (2003).

4. B. L McNaughton. C. A. Barnes. ). O'Keefe. Exp. Brain
Res. S2. 41 (1983).

5. J. O'Keefe. N. Burgess. Nature 381. 425 (1996).

6. R. U. Muller, J. L Kubie. J. B. Ranck Jr.. J. Neurosci. 7,
1935 (1987).

7. B. L McNaughton et at..}. Exp. Biol. 199, 173 (1996).

8. 0. 0. Hebb. The Organization of Behavior:

A Neuropsychological Theory (Wiley. New York. 1949).

9. E. Tulving, Elements of Episodic Memory (Clarendon,
Oxford. UK. 1983).

10. L R. Squire. PsychoL Rev. 99. 195 (1992).

11 M. Abeles. Cortkotronia: Neural Circuits oj the Cerebral
Cortex. (Cambridge Univ. Press, New York. 1991).

12. M. W. Howard. M. S. Fotedar. A. V. Datey.

M. L Hasselmo. Psychol. Rev. 112, 75 (2005).

13. W. B. Levy. A. B. HocJcing. X Wu. Neural Netw. 18. 1242
(2005).

14. L M. Frank. E. N. Brown. M. Wilson. Neuron 27. 169
(2000).

15. J. Fetbinteanu. M. L Shapiro. Neuron 40. 1227 (2003).

16. E. R. Wood. P. A. Dudchenko. R. J. Robitsek.

H. Eichenbaum. Neuron 27. 623 (2000).

17. J. A. Ainge. M. A. van der Meer. R. F. Langston.

L R. Wood. Hippocampus 17. 988 (2007).

18. W. E. Skaggs. B. L McNaughton. M. A Wilson.

C. A. Barnes. Hippocampus 6. 149 (1996).

19. 0. Dragot. G Bursaki. Neuron 50. 145 (2006).

20. J. R. Huxter, T. J. Senior, K. Allen. J. Csicsvari, Nat.

Neurosci 11. 587 (2008).

21 G Bursaki, Neuroscience 31. 551 (1989).

22. M. A Wilson. B. L McNaughton. Science 265 . 676 (1994).

23. K. Louie. M. A. Wilson, Neuron 29. 145 (2001).

24. S. A Deadwyler. T. Bunn. R. E. Hampson. ). Neurosci. 16.
354 (1996).

25. H. Eichenbaum. P. Dudchenko. E. Wood, M. Shapiro.

H. Tanila. Neuron 23. 209 (1999).

26. A Gurko. H. Hirase. J. Csicsvari, 6. Bursaki. fur. ).
Neurosci 11 . 344 (1999).

27. H. Hirase. A Gurko. J. Csicsvari, G Bursa ki. Ear. }
Neurosci 11. 4373 (1999).

28. C. Geisler, D. Robbe, M. 2ugaro. A. Sirota. G Buzsaki,
Proc. NatL Acad. Sd. U.S.A. 104 . 8149 (2007).

29. X M. Gothard. W. £. Skaggs, B. L McNaughton . }
Neurosci 16 . 8027 (1996).

30. J. O'Keefe. M L. Recce. Hippocampus 3. 317 (1993).

31 A P. Maurer. S. L Cowen. S. N. Burke. C. A Bames.

B. L. McNaughton. }. Neurosci. 26 . 13485 (2006).

32. F. SargoSni el at.. Science 312. 758 (2006).

33. B. L McNaughton. F. P. Battaglia. 0. Jensen. E. L Moser.
M. B. Moser. Nat Rev. Neurosci. 7. 663 (2006).

34. S. Fujisawa. A Amarasingham, M. T. Harrison,

G Bursa ki. Nat Neurosci. 11. 823 (2008).

35. X G Li, P. Somogyi, A Ylinen. G Buzsa'ki. /. Comp.
Neurol 339. 181 (1994).

36. G Kreiman, C. Koch. I. Fried, Nature 408 , 357 (2000).

37. J. E. Lisman. Neuron 22. 233 (1999).

38. L F. Abbott. W. G. Regehr. Nature 431 . 796 (2004).

39. M. Rabinovich, R. Huerta. G Laurent Science 321, 48
(2008).

40. M. B. Zugato, L Monconduit G. Bursa ki. Not. Neurosci.
8. 67 (2005).

41 We thank H. Hirase for sharing his data and C. Curto,

C. Geisler. S. Oren. S. Fujisawa. K. Mizuseki, A Sirota,

D. W. Sullivan, and R. L Wright for comments.
Supported by NIH (NS34994 and MH54671). NSF (SBE
0542013). the James S. McDonnell Foundation. NSF
<AA.). the Swartr Foundation (V.L). and the Robert Leet
and Clara Guthrie Patterson Trust (E.P.).

Supporting Online Material

www.scie ncema g. orgicgi/conte nt/full/32 1/58 94/13 22/DC 1

SOM Text

Figs. SI to SI 3

Table SI

Movie SI

References

29 April 2008; accepted 29 July 2008
10.112 6/science.ll59 775

www.sciencema 9 . 0 rg SCIENCE VOL 321 5 SEPTEMBER 2008

1327

Experimental Test of
Self-Shielding in Vacuum
Ultraviolet Photodissociation of CO

Subrata Chakraborty, 1 Musahid Ahmed, 2 Teresa L. Jackson, 1 Mark H. Thiemens 1 *

Self-shielding of carbon monoxide (CO) within the nebular disk has been proposed as the source
of isotopically anomalous oxygen in the solar reservoir and the source of meteoritic oxygen
isotopic compositions. A series of CO photodissociation experiments at the Advanced Light Source
show that vacuum ultraviolet (VUV) photodissociation of CO produces large wavelength-dependent
isotopic fractionation. An anomalously enriched atomic oxygen reservoir can thus be generated
through CO photodissociation without self-shielding. In the presence of optical self-shielding of
VUV light, the fractionation associated with CO dissociation dominates over self-shielding.

These results indicate the potential role of photochemistry in early solar system formation and
may help in the understanding of oxygen isotopic variations in Genesis solar-wind samples.

I sotope-selective photodissociation, or self-
shielding, is a process that occurs because of
two major parameters: (i) dissociation by
isotope-dependent spectral line absorption, and
(u) differential photolysis that depends on the
isotopic abundances. As a result, when the
spectral line corresponding to the major species
16 0 saturates, the lines corresponding to the
minor species ( 17 0, 18 0) do not, and equal

dissociation of the minor species results with
8 n O - S I8 0. CO, the most abundant oxygen-
bearing molecule in the nebula, satisfies these
criteria, and scif-shickling may occur, with equal
dissociation rates of the minor isotopes. The con-
sequence is preferential production of 17 0 and
18 0 atomic oxygen species in the region of the
disk where self-shielding is effective. Such
models assume that there is no other isotope ef

feet associated with the photolysis, a hypothesis
that has not been experimentally tested.

Isotopically selective photodissociation of CO
is invoked as an important photochemical process
in interstellar molecular clouds {1-3) to explain
the observed abundance variation of minor iso-
topomers of CO ( n C l6 0, l2 C l8 0, 12 C 17 0). CO
self-shielding has been proposed to account for
the observed meteoritic oxygen and nitrogen iso-
topic anomalies (-/). Recently, this process was
proposed to act in a very hot location near the
proto-Sun [within 0.6 astronomical unit (AU)],
and the heavy atomic oxygen ( I7 0 and 18 0) was
transported through the “X”-wind to the chondrule-
forming zone, whereas calcium-aluminum-rich
inclusions (CAls, the first condensed solids) formed
from the residual primordial nebular gas (5). To
avoid the erasure of the anomaly by isotopic ex-
change (6), it was also suggested that the self-
shielding occurred in a low-temperature prenebular
molecular cloud (7).

Another model invoked a region at low tem-
peratures (-50 K) at 30 AU from the proto-Sun
and above the disk midplane as the probable lo-
cation of CO self-shielding {8). Formation of water

'Department of Chemistry amt Biochemistry, University of
California, San Diego, la Jotta, CA 92093-0356, USA.
Chemical Sciences Division, Lawrence Berkeley National
Laboratory, 1 Cyclotron Road, Berkeley, CA 94720, USA
*To whom correspondence should be addressed. E-mail:
mthiemens@ucsd.edu

Table 1. The data obtained for the 14 experimental runs (1 through 4 and 6 through 15) grouped in 6 different experimental configurations. The
parameters used in the experiments are also displayed (see supporting online material for blank correction).

Run ft

Column

density

( 10 17

mol/cm 2 )

Exposure

time

(min)

Amount

before

fluorination

(|imol)

Amount

after

fluorination

(pmol)

N?

blank

<%>

0 2

blank

from

fluorination

(limol)

Measured Measured
S 17 0 5 18 0

(%o) <%o)

Blank

corrected

8 17 0

CW

Blank

corrected

o 18 0

(%.)

Slope

In

form:

S 17 0

(%o)

Ln

form:

6 18 0

(%o)

1 107.61 nm (11J>2 ± 0.25 tV) at room temperature (in order of increasing column density)

14

6.39

190

0.09

0.27

20.7

0.12 904.9

498.8

2136.9

1155.3

1.49

1143.2

767.9

15

10.6

510

0.25

0.45

13.0

0.14 5167.7

2430.0

8086.0

3792.6

1.41

2206.7

1567.1

2. 105.17 nm

(11.79 ± 0.25

eV) at room temperature On order of increasing

column density)

8

4.39

890

0.27

0.49

12.1

0.16 1147.1

686.2

1823.0

1081.2

142

1037.8

733.0

2

5.87

450

0.91

1.17

10.7

0.13 670.3

412.4

767.9

470.1

148

569.8

385.3

3

6.61

450

0.53

0.70

12.3

0.09 1177.9

700.2

1367.9

810.5

145

862.0

593.6

6

7.71

600

1.02

L14

13.5

1809.7

1079.3

1809.7

1079.3

141

1033.1

732.1

7

11.9

860

0.57

0.64

27.4

2445.9

1407.0

2445.9

1407.0

141

1237.2

878.4

3.

105.17 nm (11.79 t 0.25 eV) at dry-ice temperature (-66 X) (in order of increasing column density)

4

6.09

450

0.34

0.65

20.2

0.18 2031.5

1424.0

3108.1

2170.9

122

1413.0

1154.0

9*

7.54

825

0.20

35.05

0.10

34.81 43.5

43.5

5613.9

3676.4

122

1889.2

1542.5

11

9.37

450

0.17

0.41

10.6

0.20 2908.4

1905.3

6257.6

4081.9

122

1982.0

1625.7

4. 97.03 nm (12.78 t 0.25 eV) at room temperature

12

6.81

435

0.59

0.88

5.8

0.24 829.8

703.1

1161.2

978.7

112

805.4

7172

5. 94.12 nm

(13.17 ± 0.25 eV) at room temperature Cm order of increasing column density)

1

3.93

290

1.46

L54

9.9

159.7

258.8

159.7

258.8

0.64

148.1

230.2

10 f

7.03

430

0.49

0.56

9.3

0.02 7.7

117

7.6

11.3

0.67

7.6

11.3

6. 94.12 nm (13.17 ± 0.25 eV) at dry-ice temperature (-66 X)

13

6.81

435

0.67

0.72

23.9

542.9

873.7

446.1

718.6

0.69

459.6

662.1

•Air leaked through the oxygen sample tube after fluorination. Condensables are separated cryogenically and the noncondensabtes (N ? ) are separated via gas chromatography. From the
measured amount of separated the air O? amount was estimated and used for the blank correction with an air 0 2 isotopic composition of 12.1 and 23.5%» for S l, 0 and S >8 0,
respectively. fFor unknown reasons, the 5 values were quite low compared to those of a simitar experiment (#1), though the slope value (8 ,7 0/8 u 0) was the same.

1328

5 SEPTEMBER 2008 VOL 321 SCIENCE wvwv.sciencemag.org

REPORTS

icc with anomalous heavy isotopic composition
from self-shielded CO dissociation can account
for the hydrous minerals of the Allende meteorite
( 9 ). Two major assumptions were made: (i) The
initial nebular oxygen isotopic composition was
considered to be that of CAls (8 I7 0 = S l8 0 =
~40%o, and (u) no isotopic fractionation occurred
during CO photodissociation other than via the
opacity effect. We test this second assumption
through experimental isotopic measurements in
the relevant spectral region.

CO absorbs VUV photons at discrete spectral
lines, and the excited (Rydberg) states are mostly
predissociated through interaction with contin-

uum states ( 2 ). Between 90 and 1 1 0 nm, there arc
numerous (-41 ) strong absorption bands. All bands
are not equally effective for photodissociation of
CO; the lines assigned to bands with the largest
oscillator strength yield the predominant contri-
bution to the rate of photodissociation ( 2 , 10 ). The
lines at 107.61, 94. 12, 95.01, 97.03, and 94.01 ran
are most effective at lower optical depth, and at
higher optical depth ( 11 ) the lines at 95.01, 94.01,
96.89, and 92.58 nm are the most effective and con-
tribute substantially to the dissociation rate. Un-
der nebular conditions (in the presence of H 2 ,
H 2 /CO - 1(f), CO lines arc removed from the
absorption lines by H 2 and H and do not effec

Fig. 1. Oxygen isotope compo-
sition of product C0 2 in a three-
isotope plot (in logarithmic form to
incorporate the nonlinearity in
cross 8-plots for large 8 values)
showing wavelength-dependent
fractionation pattern during CO
photodissociation. The RT photolysis
products at 107.61 and 105.17 nm
[where the upper electronic state is
E'n and associated with "accidental-
predissociation" via the channel
C( 3 P) + 0( 3 P)| are extended over a
linear line with a slope value of
1.38 (with an intercept of 37.5%*),
whereas the products at -66°C of
105.17 nm are fractionated more
than the RT data and follow a relatively
lower slope of 121 (with an in-
tercept of 14.5%o). The compo-
sition of the product of 94.12 nm
[which represents a different higher-

state fn) occupancy during VUV absorption and adiabatically dissociates to form C(*D) and 0( 1 D)| at RT
follows a trend line of 0.64. The product of the same wavelength at a lower temperature (-66°0 shows
larger fractionation, as observed for 105.17 nm, but has a slope value similar to that at RT. The photolysis
product at 97.03 nm [which occupies another electronic state, C l Z, and dissociates through C( 1 D) + 0( 3 P)
channel) lies in between the results of 94.12 and 105.17 nm.

Fig. 2. Schematic dia-
gram of the CO poten-
tial energy curve, showing
the specific upper-level
occupations by different
VUV photon absorption
in this experiment The
upper-electronic state E : I1
is associated with "acciden-
tal predissociation," and
its lower vibrational states
are perturbed by another
bound triplet state, k 3 fl.

lively contribute to dissociation (lines at 91.37,
91.73, 92.87, 93.00, 93.11, 93.17, 94.63, 96.44,
98.56, 1 00.26, and 1 06.31 mn are shielded by the
absorption lines of H 2 and H). Apart from
shielding by the H 2 and H lines, some lines of
minor isotopomers (c.g., l3 CO, C 17 0, and C 18 0)
are shielded by CO lines because they lie close to
one another and, therefore, there would be no ap
preciable self-shielding at the lines at 91 .57, 91 .60,
92.28, 95.62, 97.03, 98.98, and 107.61 nm ( 11 ).
Hence, the total photodissociation rate is a com-
bination of absorptions associated with many
different transitions (table SI).

We used a windowless flow chamber with
three stages of differential pumping (fig. SI) at
the chemical dynamic bearaline (9.0.2) located at
the Advanced Light Source (ALS), Lawrence
Berkeley National Laboratory. Ultrahigh-purity
CO was passed through the cylindrical reaction
chamber and was exposed to VUV synchrotron
photons ( 10 16 photons/s) from the beamline to dis-
sociate CO to C and O atoms along the line of CO
flow. During the irradiation period, two stainless-
steel spirals at the outlet of the reaction chamber
were frozen in liquid nitrogen (LN 2 ) and collected
the final product C0 2 , formed by reaction of O
atoms and bath gas CO. The C0 2 was converted
to 0 2 by means of the fluorination method ( 12 ),
and the oxygen isotopic ratios were dcicnnincd by
isotope ratio mass spectrometry (see supporting
online material for details).

We used four different wavelengths in our
experiments— 107.61, 105.17, 97.03, and 94.12
nm — to investigate the wavelength and upper elec-
tronic excited stale-dependent isotopic fractionation
in photodissociation. At several wavelengths, we
ran the experiment at dry-ice temperatures to
evaluate possible temperature dependencies.

Experimental results, conditions, sanfile amounts
(w'ith corresponding blanks), and isotopic compo-
sitions (measured and blank -corrected values) tor
each run are presented in Table 1 , Raw data were
corrected for fluorination blanks, determined
through (measured amount) tank C0 2 fluorina-
tion of known oxygen isotopic composition. The
isotopic composition of initial CO (8 I7 0 - 25.5
and 8 18 0 = 51%*) was determined several times
by conversion to C0 2 through electric discharge
(disproportionation) of CO (complete conversion),
followed by fluorination. The heavy isotopes of O
in the product C0 2 were relatively enriched com-
pared to initial CO in a mass-independent manner
(Table 1). The extent of isotopic enrichment was
typically a few thousands per mil, depending on
experimental conditions (column density and ex-
posure time). This extent of isotopic fractionation
is large compared to other known physical or
chemical processes.

The oxygen isotopic composition of C0 2 is
shown in Fig. 1, a three-isotope plot in logarith-
mic scale [to incorporate the nonlinearity in cross
8-plot for large 8 values (7i)J. The product C0 2
was enriched in heavy O isotopes in all cases
compared to the initial CO composition, and the
resultant isotopic compositions were wavelength

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1329

REPORTS

dependent. The range of isotopic composition at
particular wavelengths derived from variation of
column densities and exposure times. All fraction-
ations, except for photolysis at 94.12 nm, were
highly mass independent The room temperature
(RT) photolysis products at 107.61 and 105.17nm
extended over a linear line with a slope of 1 .38
(with an intercept of 37%o), whereas the lower-
temperature (-66°C) products at 105. 1 7 nm were
fractioned to a greater extent than were the RT
products and follow a lower slope of 1.21 (with
an intercept of 14.5%o). The compositions of the
RT product at 94.12 nm define a trend line of
0.64 (with an intercept of 0.3%o). Results for the
same wavelength at a lower temperature (~66°Q
exhibit a larger fractionation, as observed for

105. 1 7 nm, but define a similar slope value (single
points lie over the RT trend line of 94. 1 2 nm). The
photolysis product at 97.03 nm (RT) is between
the results of 1 05. 1 7 and 94. 1 2 nm.

Self-shielding occurs as a result of optical shield-
ing of short UV light of the various isotopically
substituted CO molecules and, hence, the frac-
tionation is dependent on isotopic abundance and
not mass. The models assume that no other isotopic
fractionation is associated with photodissociation
CO self-shielding is possible through the band at

105.17 nm, but not through the band at 107.61 nm
(2), because of line broadening. In contrast, our
results show that the dissociation products for
the bands at 105.17 and 107.61 nm follow an
identical fractionation trend (slope value of 1.38).
A similar result (e.g., no self-shielding) was antic-
ipated for the line at 97.03 nm, where we observe
a mass-independent fractionated photodisso-
ciation product. Self-shielding was predicted at
the 94. 1 2-nm line; however, the experimental data
display no self-shielding effect. Hence, the dif-
ferent isotope effects during CO dissociation at
different wavelengths can derive from the nature
of higher electronic states and dissociation dy-
namics. In a schematic energy-level diagram
compiled from ( 14-19) for CO (Fig. 2), the upper
electronic state E 1 !! (a bound Rydberg state) is
associated with a type of predissociation termed
“accidental-predissociation’' (20). The lower ro-
vibronic states of E'fl arc perturbed by the
presence of another triplet state, k 3 n (21).
Predissociation from the S'!! state takes place
via the bound Rydberg k 3 n state, assisted by a
repulsive state to a near-resonance dissociative
channel C( 3 P) + 0( 3 P). In this experiment, the
vibrational energy states v - 0 and v -1 of E l n
arc populated by VUV photon absorption at
107.61 and 105.17 nm, respectively (Fig. 2), and
lead to dissociation through the “accidental-
predissociation” pathway, which proceeds via
spin-orbit coupling between singlet and triplet
states. Accidental predissociation is a well-known
near-resonance effect that occurs for narrow
ranges of rotational stales (J states). Cacciani
et al. (22, 23) showed that for v - 0 ( 1 07.62 nm),
the higher J- value states are perturbed by the k 3 !!
state, whereas lower J- value states are perturbed
for the v = 1 (105. 1 7 nm) state. The VUV source

used in our experiment is wide (full width at half
maximum - 0.25 eV), and the two bands ( 105. 1 7
and 107.62 nm) slightly overlap, but individually
populate high rotational states (J states) at RT,
which enhances the probability of accidental
predissociation. Huge isotope effects have been
observed previously as a result of near-resonance
spin -orbit coupling between singlet-triplet states.
Lira et al. (24) observed a large effect in G
isotopes during a fluorescence study of OCIO, and
Bhattacharya et al. (25) observed a large isotope
effect during C0 2 photodissociation and expert
mentally demonstrated that isotopic selective
cross-over occurs between singlet and triplet
surfaces due to spin-orbit coupling. A similar type
of anomalous isotope-dependent predissociation
in the F 3 ^ (v = 1) state of Q> has also been
observed (25). At 97.03 nm, the upper electronic
state 'l is populated via another singlet-triplet
cross-over and dissociates through the C(‘D) +
0( 3 P) channel. This singlet-triplet crossing is
different from the one associated with the E'lT
state and yields a different isotope selectivity. The
dissociation at 94.12 nm is relatively straight
forward; The 'll state is populated and adiabat-
ically dissociates through the C('D) and O('D)
channel (15). Such dissociation usually yields a
mass-dependent slope. The wavelength-dependent
fractionation pattern (Fig. 1) emphasizes that the
observed isotope effect is dominated by a physical-
chemical dissociation mechanism of the CO
molecule even in the presence of optical shielding
of the light source. A mass-independent effect in
CO 2 formation in the CO + O reaction (slope value
close to unity) has been reported (2 7). The observed
nonzero intercepts of 37.5 and 14.5%o (respectively
for 107.61 and 105.1 7 nm combined atRTand for

105.17 nm at -66 °C) as shown in Fig. 1 may be
explained through this recombination effect as a
secondary fractionation step and accounts for the
nonzero intercept with the starting isotopic com-
position, as expected for a multistage fractiona-
tion process. The temperature dependence of
fractionation is not yet clear, but may arise be-
cause the probability of a near-resonance singlet
triplet transition varies inversely with temperature,
as was seen for vibrational energy transfer between
isotopomers of CO at low temperatures (28).

A numerical analysis of absorption and dis-
sociation cross sections of different VUV lines
has quantified their differential share at the edge
of a molecular cloud in the total absorption cross
section (table SI). Van Dishoeck and Black (2)
estimated that 58% of C 18 0 dissociates through
band 31 (105.17 nm) at the center of the cloud;
however, this band contributes little to the ab-
sorption ( 1 .0%) and dissociation ( 1 .4%) at the edge
of the cloud. Other lines used in our experiment —
107.61, 97.03, and 94.12 nm — contribute ap
preciably to the absorption and dissociation.
During the exposure of a nebular gas by solar
VUV radiation, the lines listed in table S 1 are all
available for CO dissociation (assuming VUV
photodissociation of CO at the outer edge of the
disk).

The actual isotopic composition of atomic
oxygen is double that of the C0 2 that we mea-
sured, because C0 2 was produced after reaction
of an O atom with normal CO. Determination of
the O-atom compositions due to self-shielding
(calculated through photon absorption by differ-
ent isotopologs of CO with a fixed set of param-
eters, e.g., initial photon flux, column density, and
exposure time) for all experiments in Table

I yields a fractionation trend line of slope 1 in a
three-isotope oxygen plot (fig. S2). The column
density used in our experiments was optically
thicker (due to experimental constraint) than that
thought to pertain in the solar nebula and may
complicate the simple absorption calculation (fig.
S2) due to pressure broadening. Nevertheless, a
highly enriched anomalously fractionated O-atom
reservoir is possible through photodissociation of
CO at various wavelengths. Considering the low-
temperaturc data (which arc more fractionated and
slightly depleted in n O compared to RT exper-
iments), we conclude that CO dissociation at the
cold regions of the solar nebula is a potential site
for the generation of more O-atom reservoirs with
high amounts of I7 0 and ls O relative to 1<s O, but
not via self-shielding.

Water plays a vital role in defining the oxygen
isotopic distribution in the solar system as one-
third of total oxygen resides in it (9, 29-32). Recent
ly, an extremely heavy oxygen isotopic composition
(~180%o in 5 17, 18 0) of nebular water in aqueous
altered meteoritic minerals has been reported (33).
Laboratory experiments have implied that oxygen
isotope exchange between water in the gas phase
and molten silicates at high temperatures can occur
(34). Cumulative evidence suggests that there was
an anomalous oxygen isotopic reservoir. Our ex-
periments are consistent with the formation of
such a reservoir; however, they also indicate that
self-shielding is not required, thus eliminating
spatial dependencies. The results do not neces-
sarily require the initial isotopic composition of
the nebula to be S 18 0 - 8 17 0 = ~40%«, although
this composition is not ruled out

References and Notes

1. J. Bally, W. D. Ian gw. Ast/ophyi. ). 255. 143 (1982).

2. t F. van thshoeck. J. H. Black. Attrophyi ). 334. 771
(1988).

3. S. R. Federman el al.. Aslrophyi. J. 591. 986 (2003).

4. M. H. Thiemens, J. E. Heidenreich, Science 219, 1073
(1983).

5. R. N. Gaytan. Noture 415. 860 (2002).

6. 0. Navon. Ci. J. Wauetburg. forth Planet. Sd Lett 73.

1 (1985).

7. H. Yurimoto. It Kuramoto. Science 305, 1763 (2004).

8. J. R. Lyons. E. 0. Young. Noture 435. 317 (2005).

9. R. N. Gayton. Anna. Rev. Earth Planet Sd. 21. 115
(1993).

10. C. leteelter, M. Eidekberger. F. Rostas. ). Breton.

B. Thieblemont. Chen Phy%. 114. 273 (1987).

II S. Warin, J. ]. Benayoun, Y. f>. Viala. Astran. Astrophyi.
308. 535 (1996).

12. S. K. Bhattacharya, M. H. Thiemens, 2. Noturfonch. A
44a. 435 (1989).

13. E. 0. Young, G. Albert H. Nagahata. Gear hem.
Coimochim. Acta 66. 1095 (2002).

14. S. Wilson, tnl ). Quantum Chen 12, 609 (1977).

15. l. Cooper. It Kirby. ). Chem. «iyt 87. 424 (1987).

16. K. Kirby. 0. L. Cooper . ). Chem. Phyt. 90. 4895 (1989).

1330

5 SEPTEMBER 2008 VOL 321 SCIENCE wvwY.sciencemag.org

REPORTS

17. J. Baker et at.. Chem. Phys. 178. 569 (1993).

18. A. Meltinger. C. R. Vidal / Chem. Phys. 101. 104 (1994).

19. J. Baker. F. Launay. Chem Phys. Lett. 415. 296 (2005).

20. J. D. Simmons, S. G. Tilford, / Mot. Spectro sc 49. 167
(1974).

21. ). Baker. F. launay. / Mot. Spectrosc 165. 75 (1994).

22. P. Cacciani. W. Hogervorst, W. Ubachs. / Chem Phys.
102. 8308 (1995).

23. P. Cacciani et at.. Astrophys. J. 499, 1223 (1998).

24. G. Lim, S. M. Lim. S. K. Kim. Y. S. Choi. J. Chem. Phys.
111. 456 (1999).

25. S. K. Bhatlacharya. ). Savarino, M. H. Thiemens, Geophys.
Res. tut 27. 1459 (2000).

26. B. R. Lewis. S. T. Gibson, ). P. England. 6. Stark.

J. B. West / Chem. Phys. 116. 3286 (2002).

27. A. Pandey. S. K. Bhattacharya, J. Chem. Phys. 124.
234301 (2006).

28. M. L Turnidge. ). P. Reid. P. W. Barnes,

C. J. S. M. Simpson, / Chem. Phys. 108. 485 (1998).

29. E. Anders, N. Grevesse, Geodtem Cosmochim Ado 53.
197 (1989).

30. B. Fegley Jr.. Space So. Rev. 92. 177 (2000).

31. E. D. Young. Philos. Trans. R. Soc Load. A 359. 2095
(2001).

32. L 0. Young. Earth Planet So. left 262. 468 (2007).

33. N. Sakamoto et at.. Science 317, 231 (2007).

34. Y. Yu. R. H. Hewins, R. N. Clayton. T. K. Mayeda.
Geochem. Cosmochim Ado 59, 2095 (1995).

35. This work was supported by NASA grant NNX07AJ81G
under Origins and by ihe director, Office of Energy

Research, Office of Basic Energy Sciences, Chemical
Sciences Division of the U.S. Department of Eneigy under
contract DE-AC02-05CH11231. Three reviewers are
thanked for helpful suggestions that improved the
manuscript

Supporting Online Material

www.scie rtcema g.ocg/cgi /conte nt/full/32 1/58 94/13 28/DC 1

Methods

Figs. SI and S2

Table SI

References

15 April 2008; accepted 11 July 2008
10.112 6/science.ll59178

Identification of Active Gold
Nanoclusters on Iron Oxide
Supports for CO Oxidation

Andrew A. Herzing, t2 Christopher J. Kiely, 1 * Albert F. Carley, 3
Philip Landon, 3 Graham ]. Hutchings 3 *

Gold nanocrystals absorbed on metal oxides have exceptional properties in oxidation catalysis,
including the oxidation of carbon monoxide at ambient temperatures, but the identification
of the active catalytic gold species among the many present on real catalysts is challenging. We
have used aberration-corrected scanning transmission electron microscopy to analyze several iron
oxide-supported catalyst samples, ranging from those with little or no activity to others with
high activities. High catalytic activity for carbon monoxide oxidation is correlated with the presence
of bilayer clusters that are ~0.5 nanometer in diameter and contain only ~10 gold atoms.

The activity of these bilayer clusters is consistent with that demonstrated previously with the use of
model catalyst systems.

G old nanoparticles on oxide supports are
being used to catalyze an increasing
number of reactions {1-6), but there is
continuing interest in the relatively simple low-
temperature oxidation of CO {1-3, 7), especially
concerning the nature of the active species. Some
proposals emphasize the size and morphology of
the gold particle and its interface with the support
{1, 8), whereas others postulate that the metal
oxidation state (9, 1(f) and support material can
have strong effects {1, 11). Most of the tundamen-
tal insights into this reaction have come from well-
defined model studies {12-1S) in which specific
structures are tailor-made {14). However, the
difficult)' lies in linking these informative model
studies to those carried out on real catalysts, which
are inherently more complex. Here, we report a
microscopy study of an Au/FcO x catalyst and
demonstrate that, among a number of gold nano-
structures present, the origin of the activity is
uniquely with gold bilayer nano-

’Center for Advanced Materials and Nanotechnology, Lehigh
Univen'rty, 5 East Packer Avenue, Bethlehem, PA 18015-3195,
USA. National Institute of Standards and Technology, Surface
and Microanalysis Science Division, 100 Bureau Drive. Mailstop
8371, Gailhersburg, MD 20899-8371, USA. ’Cardiff Catalysis
Institute School of Cheinistiy, Cardiff University, Cardiff, CF10
3 AT, UK.

•To whom correspondence should be addressed. E-mail:
chk5@lehigh.edu (C.JX); hutch@cardiff.ac.uk (G.J.H.)

clusters that are -0.5 nm in diameter and con-
tain -10 Au atoms.

Gold supported on FeOOH is a highly active
catalyst for the low-teraperature oxidation of CXD
{16), and we recently reported a link between
catalyst performance and catalyst drying condi-
tions (10). A pair of 2.9 atomic % Au/FeO* samples
(denoted samples 1 and 2) were derived from the
same coprecipitated precursor. Sample 1 was dried
in a tube furnace (static air furnace was ramped
after insertion at 1 5°C/min to 120°C; 16 hours),
whereas sample 2 was dried in a gas chromatog-

raphy (GC) oven (flowing air; sample inserted
into the furnace at 120°C; 16 hours). The Au
loading in each was identical, and the underlying
disordered FeOOH supports had similar surface
areas (~1 90 m 2 g -1 ). X-ray energy-dispersive spec-
troscopy (XEDS) analysis and high -angle annular
dark-field (HAADF) imaging experiments (10)
indicated that both samples contained 2 to 15 -nm
Au particles, with mean particle sizes of 5.4 nm
for sample 1 and 7.0 nm for sample 2.

If the CO oxidation activity was dictated solely
by the Au particle size, as some researchers have
suggested (1, 8), then these two samples should
have exhibited similar activities. In feet, sample
1 would probably be expected to have a slightly
better performance by virtue of the Au particles
being slightly smaller, on average. However, catalyt-
ic testing of these two samples under standard con-
ditions (total flow rate of 66,000 h 0.5 volume %
CO) revealed feat sample 2 achieved 100% CO
conversion at 25°C, whereas sample 1 gave only
trace CO conversion (<1%). Even at much higher
flow rates, sample 2 remained active (i.e., at a total
flow rate of 450,000 vol.gas/vol.catalysth: the
conversion was 25%), whereas sample 1 was to-
tally inactive Previously {10), we were unable to
determine the origin of this marked effect because
of the resolution and sensitivity limitations of the
characterization techniques available. We consider
that at least two possibilities exist to explain this
difference: (i) There could be highly dispersed Au
species present in sample 2 that were beyond the
resolution limit of the conventional HAADF and

Fig. L Low-magnification aberration -corrected HAADF-STEM images from (A) the inactive, tube
furnace catalyst (sample 1) and (B) the highly active, GC oven catalyst (sample 2). Au particle size
distribution in both samples appears to be very similar at this magnification.

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1331

REPORTS

scanning transmission electron microscopy
(STEM)-XEDS techniques used previously, or
(ii) the active catalyst contained some form of
cationic Au species that was absent in the inactive
sample. Three two possibilities need not be
mutually exclusive.

To investigate this unexpected difference in
reactivity, we reexamined the samples using a
state-of-the-art JEOL 2200FS in STEM mode
equipped with a double-hexapole spherical aberra-
tion corrector manufactured by CEOS GmbH ( 17 ).
The improvement to tire HAADF image resolution,
from the use of a 0.1 -nm diameter aberration-
corrected STEM probe containing 50 pA of
current, makes this an ideal instrument to tret
whether there were Au species present in the
highly active sample that were not detected
earlier ( 10 ). Previously, this method has success-
fully been used to image atomic dispersions in the
La/AliOj, Pl/Al 2 0 3 , and Au/Ti0 2 systems ( 18 ).
The lower-magnification HAADF images for each
of the dried Au/FeO* catalysts (Fig. 1, A and B)
confirm the earlier results, with particles between 2
and 15 nm in diameter. However, at higher magni-
fication (Fig. 2, A to D, and figs. SI and S2), die
actual Au particle size distribution and morphology
in these samples are quite different. Both samples
contain larger (2 to 15 nm) Au particles and a
considerable number of individual Au atoms
(indicated by white circles) dispersed on the iron
oxide surface.

The observation of these individual atoms
ensures that we are now resolving all of the types
of Au species present. However, in addition a
large population of subnanometer Au clusters
was found in the active sample 2 (circled in black
in Fig. 2, C and D) (see also fig. SI ) dial was not
detected in the inactive sample. These Au clusters,
the majority of which were 0.2 to 0.5 nm in
diameter, contain at most only a few Au atoms.
The contrast level exhibited by the 0.2- to 0.3-nm
clusters, which are predominant in Fig. 2D, is sim-
ilar to that of the individual Au atoms, implying that
they are monolayer structures. In comparison, the
0.5-nm dusters highlighted in Fig. 2C exhibit
markedly higher contrast in the HAADF image than
that of the 0.2- to 0.3-nm clusters, suggesting that
these larger dusters are most likely bilayers of Au.

Determination of the exact number of atoms
in these various structures is complicated by sev
eral factors, including the slight contraction of the
Au-Au bond distance known to occur in particles
of this size ( 19 ). However, a rough estimate is still
informative, and biiaycrs of ~0.5-nm lateral di-
mension would contain ~10 atoms (fig. S3A). In
comparison, the 0.2- to 0.3-nra clusters of similar
contrast to monolayers could contain about three to
tour atoms (fig. S3B). In contrast, at estimate
based on the volumetric packing density of Au
(—59 &oms/nm 5 ) suggests that the larger nano-
particles (5 to 7 nm) would contain 1900 to 5250
atoms if they have a hemispherical geometry. Thus,
the subnanometer clusters and individual atoms
observed in the highly active sample represent
only a very minor fraction of the total Au content

We used atomic-resolution STEM imaging to
estimate the number fraction of the total Au
particles represented by both clusters and nano-
particles by surveying several support areas that
had -100 larger nanoparticles, along with their
neighboring vicinities ( 17 ). In terms of their fre-
quency of observation, the individually adsorbed
Au atoms represented 44 1 4.5% of the Au species,
whereas 19 ± 3.4% were 0.2- to 0.3-nm clusters,
1 8 ± 3.0% were 03- to 0.5-nm clusters, and 19 t
3.5% were particles >1 nm (-50% were >5 nm).

However, considering the number of Au atoms
in each of these structures (as discussed above)
suggests that the individual atoms represent only
0.13 ± 0.07 atomic % and that the monolayer and
bilayer subnanometer clusters combined represent
only 1.05 ± 0.72 atomic % of the total Au loading
( 18 ), with the remaining 98.82 ± 0.80 atomic % of
Au contained in the larger particles. Because of the
very large disparity in the number of atoms they
contain, the number of subnanometer clusters would
have to exceed that of the 5-nm particles by a mar
gin of nearly tour to one for them to represent just
2 atomic % of the total Au loading. Therefore, it
is probable that these minority Au species would
not be easily detected with traditional “bulk”
techniques such as extended x-ray absorption fine
structure or Mossbauer spectroscopy, or even by
surface analysis techniques such as x-ray photo-

electron spectroscopy (XPS), because their con-
tribution to the total signal would be minimal
compared with that of the larger nanoparticles ( 17 ).

The statistical accuracy of the current HAADF
analysis must be considered, because undercount-
ing of the subnanometer Au clusters may result
from the nature of the STEM imaging process in
an aberration-corrected instrument. The ability to
detect single atoms on an oxide is highly depen-
dent on the vertical position of the focused elec-
tron probe relative to the surface ( 20 ). Because
adsorbed atoms can be present on either the top
or bottom surface of the underlying oxide par-
ticle, imaging witli single-atom resolution is pos-
sible only when one of the surfaces is exactly in
focus. Practically, this limitation means that in
any given image, single atoms are probably only
detected on one surface of the support particle,
and atoms on the opposite side may not be visi-
ble, whereas larger nanoparticlcs can be detected
on either side of the support over a much wider
range of defocus values. In the intermediate case
of 0.2- to 0.5-nm clusters, it is not clear if the same
narrow dcfocus range required to resolve individ-
ual atoms is necessary. However; even if it is pes-
simistically assumed that our analysis lias detected
only 50% of the total number of all the individual
Au atoms, the atomically dispersed Au would still
represent only 61 ± 3.6% of the total number of

Fig. 2. High-magnification aberration-corrected STEM-HAADF images of (A and B) the inactive
(sample 1) and (C and D) the active (sample 2) Au/FeO, catalysts acquired with the aberration-
corrected JEOL 2200FS. The white circles indicate the presence of individual Au atoms, whereas the
black circles indicate subnanometer Au clusters consisting of only a few atoms. Note the presence
and image intensity difference of two distinct cluster-types: In (O there are 0.5 nm higher-contrast
clusters, whereas in (D) 0.2- to 0.3-nm low-contrast clusters dominate. This difference indicates
that bilayer and monolayer subnanometer Au clusters are present in the active catalyst.

1332

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

REPORTS

Au species, or -0.25 ± 0.2 atomic % of the total
Au loading.

The observation that the active species in our
Au/FcO x catalysts consist of subnanometer clus-
ters differs from numerous earlier investigations
that identified 2- to 5-nm particles as the critical
“nanostructure.” However, a number of recent
experimental and theoretical ( 21-23) studies have
suggested that the CO oxidation activity should
continue to increase as the Au particle size ap-
proaches 1 nm because of the presence of a larger
fraction of Au atoms with low coordination in the
smaller particles. Low-coordination Au atoms
possess a d band that is closer to the Fermi level
than their close-packed counterparts, so they can
adsorb 0 2 molecules more readily. More specif
ically, Falsig et al. predicted that adsorption
energies for CO and O are ideally suited tor max
imizing the reaction rate of CO oxidation at sixfold
coordinated Au comer atoms, whereas the adsorp
tion energies at close-packed Au faces are inferior
for this purpose ( 23 ). Because clusters smaller than
1 nm should have an even greater fraction of low-
coordinated Au atoms, reaching -90% as a cluster
size of 0.5 nm is approached, the role of sub-
nanometer clusters may be crucial Rashkeev et al.
(24) have recently presented HAADF evidence that
subnanometer mono-, bi-, and trilayer Au clusters
coexisted in active Au/Tt0 2 catalysts, although they
were unable to isolate the relative importance of
each of these subnanometer species and did not re-
port any spectroscopic measurements to determine
the possible role of substrate surface chemistry.

Scanning tunneling microscopy studies on
model catalysts by Matthey el al. (15) showed
that subtle changes in the chemistry' of an oxide
surface could alter the energetically favorable Au

structures that it stabilized and supported. Specif
ically, an oxygen-rich Ti0 2 (110) surface could
stabilize a range of Au species from one to seven
atoms in size. In contrast, single Au atoms and Au
trimers were the only stable configurations on a
reduced TiC^ (110) surface, whereas weakly banded
two-dimensional Au raftlike species were tire only
stable structures on a stoichiometric TiOj (110)
surface. Similar oxygen-rich surfaces to those re-
ported for Tj0 2 have also been identified for FeO x
(25); however, to our knowledge, systematic
surface science studies and first-principle calcu-
lations of the stable Au structures present on
FeOOH or F^Oj have not yet been performed.

On the basis of the observations of the two
dried catalysts alone, we are unable to definitive-
ly comment on whether the bilayer 0.5 -run clus-
ters, the monolayer 0.2- to 0.3-nm clusters, or both
arc responsible for the high activity observed or
whether some subtle substrate chemistry effect is
coming into play. Although individual Au atoms
and larger (3 to 5 nm) particles are stable on both
catalysts, the subnanometer Au clusters were only
stable on the surface of catalyst 2 dried under a
flowing air. The critical role played by such slight
changes in the catalyst preparation route may help
to explain the sometimes radically differing
activities reported. We have used XPS to probe
the nature of the surface species present in both
samples. Analysis of the Au(4f) spectra (fig. S4. 1)
showed that, in both catalysts, the Au(4f 7(2 )
binding energy was 85.1 eV, characteristic of
Au* species. The signal from sample 1 shows some
broadening to higher binding energy that probably
arose from a small amount of Au , most likely
present as AuOOH. The presence of hydroxylated
Au species in this inactive sample suggests that the

lack of air circulation during the drying process
inhibits the removal of the hydroxyl species from
the catalyst surface, whereas this process occurs
more efficiently under flowing air conditions.

This explanation was confirmed by the O(ls)
and C(ls) spectra from these samples (figs. S4.2
and S4.3, respectively). Both catalysts exhibit a
main O(ls) component characteristic of oxidic
oxygen (O 2 ) at 530.4 eV, together with a shoulder
at 531.8 eV. The latter feature originates from a
combination of hydroxyl and carbonate species
and is much more pronounced in sample 1. The
G(ls) spectra, which show clear evidence of car-
bonate species, are similar for both. Therefore, the
higher intensity of the O(ls) shoulder at 531.8 eV
in the inactive catalyst 1 arises from the increased
presence of hydroxyl groups relative to the active
catalyst 2 dried in flowing air. This higher degree
of hydroxylation and reduction of sample 1 seems
to enhance the ability of the subnanometer clusters
to sinter into larger particles and thus deactivate the
catalyst Hence, on the basis of the evidence pre-
sented so far, we are unable to determine whether
it is the presence of subnanometer clusters or the
differing degree of hydroxylation of the support
that is controlling the activity.

Three separate portions of the highly active cat-
alyst 2, denoted 3, 4, and 5, were calcined (3 hours
at 400°, 550°, and 600 °C, respectively). These
heat treatments converted the support material to
crystalline FC2O3 and progressively reduced the
surface area (catalysts 2: 190 m 2 g -1 , 3: 45 m 2 g~ 1,
4: 16 ra 2 g“\ and 5: 11m 2 g -1 ). Atomic ab-
sorption spectroscopy confirmed that all of the
Au (2.9 atomic %) was retained. CO conversion
decreased from 100% in the fresh, uncalcined state
to 91% (400°C), 3 1% (550°C), and <1% (600°Q.

Fig. 3. Aberration -corrected
STEM-HAADF images of the
active 2.9 atomic % Au/FeO x
catalyst 2 calcined for 3 hours
at (A and B) 40<rC (sample 3),
(C and D) 550°C (sample 4),
and (E and F) 600°C (sample
5). The heat-treatment pro-
cedures have substantially
decreased the population of
subnanometer Au clusters
relative to the highly active,
dried catalyst, while at the
same time they have increased
the population of particles in
the 1- to 3-nm range.

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1333

REPORTS

Aberration -corrected STEM-HAADF images
(Fig. 3, A to F, and figs. S5.1 to 5.3) revealed that,
although substantial particle sintering had occurred
in all calcined catalysts, each sample also retained a
population of the subnanometer clusters and in-
dividual Au atoms previously seen in the dried
catalyst that had persisted through the various
calcination treatments.

The number frequency of the various gold species
encountered in samples 2, 3, 4, and 5 is sum-
marized in Fig. 4. The number traction of atom-
ically dispersed species drops from -44% to 35%
across this sample set, whereas, simultaneously,
the number fraction of particles greater than 1 mn
progressively increases from 19 to 35%. However,
our previous deductions from comparing samples
1 and 2 indicate that the supported subnanometer
cluster species (and not the dispersed Au atoms or
>1 -nm particles) are active for CD conversion. It is
also apparent from Fig. 4 that the number pop-
ulation of monolayer clusters remains relatively

60%

constant at -19%, whereas the bilaycr cluster fre-
quency gradually drops from 18% to <5% across
this same sample set. Hence, the dramatic decrease
in catalytic activity exhibited by samples 3, 4 and
5 can be directly correlated to the marked decrease
in the number density of the 0.5-nm bilayer
clusters as a result of their sintering into less active
1 - to 2-nm Au particles.

XPS of samples 3, 4, and 5 (fig. S6) showed
no notable difference in either of the Au(4f),
Fe(3s), or Fc(2p) profiles (fig. S6.1 and 6.2) as-
sociated with this series of calcined catalysts.
The O(ls) signals (fig. S6.3) are also all similar,
with a high-binding energy tail indicating the
presence of some residual hydroxyl and carbonate
species, but these are very different from the dried
sample (spectra shown in Fig. 5). The presence of
carbonate species is confirmed by the weak, high
binding feature in the C(ls) spectra (fig. S6.4).
This feature is broader for sample 5, possibly re
fleeting carbonate species in distinct chemical envi-

■ i- ad KN (100%)
m Calcined 400 C (91%)
□ Calane<J550C(3l%)

« fi/Wl C I* « W \

romnents. and is consistent with small differences
in the high-binding energy tail in the 0(1 s) spec-
trum tor this sample.

To determine whether the small differences in
samples 3 and 4 and the larger differences with
the other samples correlate with the catalytic per
tonnance, we deconvolved the O Is signals into
oxide and the higher-binding energy hydroxyl
and carbonate species to obtain the relative frac-
tions of each species present. However, the data
for either the relative area percent of the higher-
binding energy species (samples 2: 60%, 3: 28%,
4: 34%, and 5: 47%) or the hydroxyl/oxide ratio
(samples 2: 1 .5, 3: 0.4, 4: 0.5, and 5: 0.9) do not
correlate with the observed activity trend. Thus,
the similarity in the surface chemistry of samples
3 and 4, in particular, and the Au oxidation state
in these calcined samples strongly imply that the
progressive decrease in catalytic activity upon sin-
tering is attributable to the gradual decrease in the
population of the subnanometer bilayer clusters of
gold.

Theoretical and model studies (23, 26-29) have
shown that a critical factor in the catalytic activity
of Au is the ability of the clusters to simultaneously
adsorb both reactant molecules. Yoon et al. and
Hakkinen et al. (27, 28) showed that the smallest
Au cluster on MgO known to be active for CO
oxidation is an oc tamer. Our observations arc con-
sistent with these theoretical studies, because the
active bilaycr subnanometcr clusters in our system
contain -10 atoms, whereas the monolayer clus-
ters (which have only three to four atoms) and the
isolated Au atoms appear to be essentially inactive
for CO oxidation.

Finally, CO oxidation activity of model Au cat-
alysts on Ti0 2 surfaces is maximized when the Au
structures are two atomic layers thick (14, 28, 29),
resulting in turnover frequencies (TOFs) as high
as 3 .7 s -1 . In the present case, the TOF of catalyst
2 exhibiting 100% CO conversion was initially
calculated to be 0.016 s' 1 under standard conditions
and 0.027 s -1 at fee higher flow rate where total
conversion (which is a more reliable estimate of tire
catalyst activity) was not observed. However, if the
TOF is recalculated assuming that tire bilayer clus-
ters are the only active species and using a rea-
sonable estimate of the fraction of the total Au that
they contain (0.6 atomic %), the result is a TOF of

2.7 s" 1 at the standard conditions and 3.5 s -1 at the
higher-flow rate condition. These re-estimated
TOF values arc reasonably similar to the value of

3.7 s -1 of the model Au/Ti0 2 catalyst (14).

These studies describe the full range of active

and inactive Au species that are present within
supported Au/FeO x and Au/Fe 2 0 3 catalysts. Sub-
tle changes in sample-preparation route and cal
cination temperature can influence tire formation,
stability, and relative population of these various
Au species. Herein, we have reported that highly
active subnanometer Au clusters can be synthe-
sized writh a traditional chemical preparation route.
Although the highest-activity catalysts correspond
to uncalcined materials, experiments involving sys-
tematic calcination treatments have allow'ed us to

Atoms

Monolayers Bilayers

>1 nm

Fig. 4. Relative populations of (i) dispersed Au atoms, (ii) 0.2- to 0.3-nm monolayer Au clusters,
(iii) 0.5-nm bilayer Au clusters, and (iv) Au nanoparticles >1 nm in diameter, as a function of
catalyst calcination temperature and measured CO conversion. The error bars correspond to two
SDs on the size measurements.

Fig. 5. 0(ls) photoemission
spectra from the Au/Fe ? 0 3 catalysts
(A) dried at 120°C and then cal-
cined at (B) 400°C (C) 550°C, and
(D) 600°C.

1334

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

REPORTS

deduce that it is primarily the 0.5 -nm bilaycr clus-
ters, rather titan 0.2- to 0.3-nm monolayer Au
clusters, that are active for CO oxidation on FeO*
supports.

References and Notes

1. M. Hawta, CATTECH 6. 102 <2002).

2. M. Haiuta, Gold Bull. 37. 27 (2004).

3. A S. K. Hash nr., 6. ]. Hutchings. Ange tv. Chem. InL Ed.

45, 7896 (2006).

4. M. D. Hughes et a!.. Nature 437. 1132 (2005).

5. D. I. Enache et at.. Serene 311, 362 (2006).

6. A. Conra. P. Serna. Science 313, 332 (2006).

7. J. Guzman, S. Carrettin. A. Conra . ). Am. Chem. Soc.

127. 3286 (2005).

8. ). J. Pietron. R. M. Stroud. D. R. Roll son. Nano Lett 2.

545 (2002).

9. J. Guzman, B. C. Gates, jt Am. Chem. Soc 126. 2672

(2004).

R ecent cosmological observations suggest
that the universe’s expansion is accelerat-
ing. Several lines of evidence corroborate
this, including results from distant supemovae (1, 2\
the cosmic microwave background (5), and the
clustering of matter ( 4 , 5). However, the current
observations are all essentially geometric in nature,
in that they map out space, its curvature, and its
evolution. In contrast, a direct and dynamical de-
termination of the universe’s expansion history is
possible by observing the slow drift of cosnolog-
ical redshifts, z, that is inevitable in any evolving
universe (6). No particular cosmological model or
theory of gravity would be needed to interpret die
results of such an experiment However, the cos-
mological redshift drift is exceedingly small and

’Max-Planck Institut fur Quantenoptik, Hans-Kopfcrmann-
Strasse 1, 0-85743 Garchirg, Germany. 2 Menlo Systems GmbH,
Am Klcplerspitz 19, D-8Z152 Martinsried, Germany. ’European
Southern Observatory, Kart-Sch warzsch i Id -Strasse 3, D-85748
Garching, Germany. Tentre for Astrophysics and Supercomput
ing, Swinburne University of Technology, Wail H39, Post Office
Box 218, Victoria 3122, Australia. 5 <iepenheuer-lnstitut fur
Sonnenphysik, Schdneckstr. 6, D-79104 Freiburg, Germany.
•To whom correspondence should be addressed. E-mail:
thu@mpq.mpg.de

10. G. J. Hutchings et al.. ). CatoL 242. 71 (2006).

11. M. M. Schubert et at.. ). Catal. 197. 113 (2001).

12. R. Meyer, C. Lemire. Sh. K. Shaikhutdinov. H.-J. Freund.
Gold Bull. 37. 72 (2004).

13. M. Valden. X. Ui. D. W. Goodman. Science 281. 1647
(1998).

14. U. S. Chen. D. W. Goodman. Science 306. 252 (2004).
published online 26 August 2004; 10.1126/
science. 1102420

15. D. Watthey ef at.. Science 315. 1692 (2007).

16. N. A. Hodge et at.. Catal. Today 72. 133 (2002).

17. Materials and methods are available as supporting
material on Science Online.

18. M. Varela ef al.. Annu. Rev. Mater. Re s. 35. 539 (2005).

19. C. W. Mays ef al.. Surf. Sci. 12, 134 (1968).

20. S. Wang ef at.. Nat. Mater. 3. 143 (2004).

21. T. V. W. Janssens ef at.. Tap. CataL 44. 15 (2007).

22. W. Yan ef at.. /. Rhys. Chem. 8 109. 10676 (2005).

23. H. Falsi g ef at.. Ange w. Chem. InL Ed. AT. 4835
(2008).

difficult to measure; for currently favored models
of the universe, with a cosmological constant pa-
rametrizing the acceleration, the redshifts of objects
drift by less than ~1 cm s -1 year" 1 (depending on
their redshifts).

Nevertheless, the suggestion that the so-called
Lyman-a “forest” seen in high-redshift quasar
spectra is the best target for this experiment (7)
was recently supported by cosmological hydro-
dynamical simulations (8). The forest of absorption
lines is caused by the Lyman-a transition arising in
neutral hydrogen gas clouds at different redshifts
along the quasar sight-lines. Detailed calculations
with simulated quasar spectra show that the
planned 42-m European Extremely Laigc Tele-
scope (E-ELT), equipped with the proposed
Cosmic Dynamics Experiment (CODEX) spectro-
graph ( 9 ), could detect the redshift drift con-
vincingly with 4000 hours of observing time over
a ~20-ycar period (8). Therefore, as the observation
is feasible (in principle), overcoming the many
other practical challenges in such a measurement is
imperative. Important astrophysical and technical
requirements have been considered in detail, and
most are not difficult to surmount (8, 10). One (but
not the only) extremely important requirement is

24. S. Rashkeev ef at.. Rhys. Rev B 76, 035438 (2007).

25. A. Bongiomo, U. landman. Rhys. Rev. Lett 95. 106102
(2005).

26. M. Abu Haija ef aL Surf. Sci. 600. 1497 (2006).

27. 3 Yoon et aU Science 307. 403 (2005).

28. H. Hakkinen et at. Angevt. Chem. kit. Ed. 42, 1297
(2003).

29. V. A. Bondzie ef at.. Catal. Lett 63. 143 (1999).

30. We thank the Athena project of the Engineering and
Physical Sciences Research Council. NSF, NASA and the
National Research Council Postdoctoral Associate
program for funding this work.

Supporting Online Material

www.scie5cemag.otg/cgi/content/fud/321/5894/1331/DCl
Materials and Methods
Figs. SI to S6

25 April 2008; accepted 23 July 2008
10.112 6/science.ll59639

that the astronomical spectrographs involved must
have their wavelength scales calibrated accurately
enough to record -1 cm s~ l velocity shifts (-25 -kHz
frequency shifts) in the optical range. Moreover,
this accuracy must be repeatable over ~20 year
time scales.

Although the redshift drift experiment requires
demanding precision and repeatability, precisely
calibrated astronomical spectrographs have several
other important applications. For example, Jupiter-
and Ncptunc-mass extrasolar planets have been
discovered by the reflex Doppler motion of their
host stars (11-13), but detecting Earth mass planets
around solar-mass stars will require -5 cm s“‘
precision maintained over several-year time scales
(14). Another example is the search for shifts in
narrow quasar absorption lines caused by cosmo-
logical variations in the fundamental constants of
nature (15-17). Recent measurements (18-21)
achieve precisions of -20 m s~\ but the possibility
of hidden systematic effects, and the increased
photon-collecting power of future ELTs, warrant
much more precise and accurate calibration over
the widest possible wavelength range.

Laser frequency combs (LFCs) offer a solu-
tion because they provide an absolute, repeatable
wavelength scale defined by a series of laser modes
equally spaced across the spectrum. The train of
femtosecond pulses from a mode-locked laser oc-
curs at the pulse repetition rate, f [ep , governed by
the adjustable laser cavity length. In the frequency
domain, this yields a spectrum, f n =f xo + n * f—,
with modes enumerated by an integer n~ 10 5 to l(r.
The carrier envelope offset frequency, <f„ p,
accounts for the laser's internal dispersion, which
causes the group and phase velocities of the pulses
to differ (22). Thanks to the large integer n, the op-
tical frequencies f„ are at hundreds of THz whereas
both f iep aadfceo are radio frequencies and can be
handled with simple electronics and stabilized by
an atomic clock (22). Each mode’s absolute fre-
quency is known to a precision limited only by
the accuracy of the clock. Even low-cost, portable
atomic clocks provide ~1 cm s _! (or 3 parts in I0 n )
precisioa Because LFC light power is much higher
than required, the calibration precision possible is
therefore limited by the maximum signal to noise

Laser Frequency Combs for
Astronomical Observations

Tilo Steinmetz, 1 - 2 Tobias Wilken, 1 Constanza Araujo-Hauck, 3 Ronald Holzwarth, 1 - 2
Theodor W. Hansch, 1 Luca Pasquini, 3 Antonio Manescau, 3 Sandro D'Odorico, 3
Michael T. Murphy/ Thomas Kentischer, 5 Wolfgang Schmidt, 5 Thomas Udem 1 *

A direct measurement of the universe's expansion history could be made by observing in real time
the evolution of the cosmological redshift of distant objects. However, this would require
measurements of Doppler velocity drifts of ~1 centimeter per second per year, and astronomical
spectrographs have not yet been calibrated to this tolerance. We demonstrated the first use of a
laser frequency comb for wavelength calibration of an astronomical telescope. Even with a simple
analysis, absolute calibration is achieved with an equivalent Doppler precision of ~9 meters per
second at -1.5 micrometers — beyond state-of-the-art accuracy. We show that tracking complex,
time-varying systematic effects in the spectrograph and detector system is a particular advantage
of laser frequency comb calibration. This technique promises an effective means for modeling and
removal of such systematic effects to the accuracy required by future experiments to see direct
evidence of the universe's putative acceleration.

www.sciencernag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1335

REPORTS

ratio (SNR) achievable with the detector. For mod
cm astronomical charge-coupled devices (CCDs),
the maximum SNR in a single exposure is limited
by their dynamic range but is still sufficient to
achieve -l eras' 1 precision (23). Furthermore,
because LFC calibration is absolute, spectra from
different epochs, or even different telescopes, can
be meaningfully compared.

The main challenge in reaching ~1 cm s' 1 cal-
ibration accuracy will be the measurement and,
eventually, mitigation and/or modeling and re-
moval of systematic effects in astronomical spectro-
graphs and detectors. For typical high-resolution
spectrographs, a -1 cm s' 1 shift corresponds
roughly to the physical size of a silicon atom in
the CCD substrate. Only with the statistics of a
very large number of calibration lines can the re-
quired sensitivity be achieved, provided that sys-
tematic effects can be controlled accordingly (10).
For example, even in a highly stabilized, vacuum-
sealed spectrograph, small mechanical drifts will
slightly shift the spectrum across the CCD. Al-
though this can easily be tracked to first order,
other effects such as CCD intrapixcl sensitivity
variations will be important for higher precision.
Discovering, understanding, and eventually model-
ing and removing these effects is crucial for the
long-term goal of accurate calibration; tests of
LFCs on astronomical telescopes, spectrographs,
and detectors are therefore imperative.

We have conducted an astronomical LFC test
on the German Vacuum Tower Telescope (24)
(VTT) (Fig. 1). We used a portable rubidium
clock with a modest accuracy of 5 parts in 10 u
(or 1.5 cm s' 1 ); much more accurate clocks are
available if needed. This sets the absolute un-
certainty on the frequency of any given comb mode.
The VTT can be operated at near-infrared wave-
lengths, thereby allowing a relatively simple and
reliable fiber-based LFC to be used. The erbium
doped fiber LFC used had = 250 MHz which,
despite the VTT spectrograph having higher resolv-
ing power (resolution of 0.8 GHz or 1.2 km s' 1 )
than most astronomical spectrographs, is too low
for modes to be resolved apart. Filtering out un-
wanted modes by using a Fabry-Perot cavity (FPC)
outside the laser (25, 26) was suggested as one
solution (23, 27) and has proven effective (28, 29).
The FPC comprises two mirrors separated by a
distance smaller than the laser cavity length so that
all modes, except every /nth (m > 1 ), are inter
fcromctrically suppressed (Fig. 1, lower panel).
We used a FPC stabilized to a filter ratio, m, by
controlling its length with an electronic servo sys-
tem to generate effective mode spacings, m x f^,
between 1 and 15 GHz. The degree to which the
unwanted modes arc suppressed is an important
parameter; The FPC transmission function falls
sharply away from the transmitted mode frequencies
but, because nearby suppressed modes are not
resolved from the transmitted ones by the spec-
trograph, small asymmetries in this function (espe-
cially combined with time variations) can cause
systematic shifts in the measured line positions.
With our setup, we achieve an unwanted mode sup-

pression of more than 46 dB at filter ratios m < 20.
Other possible systematic shifts due to the filtering
have been identified (29) aid need to be controlled

LFC spectra were recorded with and without
the spectrum of a small section of the Sun's photo-
sphere at wavelengths ~1 .5 pm. A sample m =
15-GHz recording, superimposed with Fraunhofer
and atmospheric lines, is shown in Fig. 2. To es-
timate our calibration accuracy and to test the
spectrograph's stability, we analyzed several expo-
sures of the LFC only. Individual Lorentzian
functions were fitted to the recorded modes as a
function of pixel position and identified with the
absolute comb frequencies,/,, which are referenced
to the atomic clock (10). The dense grid of modes
allows die spectrograph's calibration function (Fig.
3A) to be determined to very high accuracy; even
a simple, second-order polynomial fit to the pixel
versus-frequency distribution has only 9 m s' 1
root mean square (RMS) residual deviations around
it (Fig. 3B), and this remains almost unchanged
with higher-order polynomial modeling (10).

With traditional calibration techniques, such
as thorium comparison lamps, I 2 gas absorption
cells or Earth’s atmospheric absorption lines for
calibration achieve ~10 m s'' absolute precision
per calibration line at best (30). Thus, even with
these “first light” comb recordings, we already
demonstrate superior absolute calibration accura-
cy. Because more than 1 0 4 modes will be available

in a larger-bandwidlh LFC, the statistical unoer
tainty would be reduced to the 1 cm s' 1 regime if
the residuals were truly random. However, the
theoretical shot noise limit calculated from the
number of photons recorded per comb mode is
much smaller than 9 m s _l ; systematic effects from
the spectrograph and detector system evidently
completely dominate the residuals.

The main reason for testing LFCs at real tele-
scopes, on real astronomical spectrograph and de-
tector systems, is to understand how to measure
and then mitigate and/or model and remove such
systcmatics. Because the VTT spectrograph is
not stabilized (i.e., temperature-, pressure- and
vibration-isolated), instrument drifts are expected
and the VTT LFC spectra can already be used to
track them accurately. From a time scries of ex-
posures, we derive a drift in the spectrograph of
typically 8 m s' 1 min' 1 (5 MHz min" 1 ) (10).
Much lower drift rates have been demonstrated
with suitably stabilized instruments [e.g., ~1 m s' 1
over months with HARPS (7i)); although the
VTT is not optimized for stability, this does not
affect its usefulness to test calibration procedures.
Indeed, different modes are observed to drift at
different rates, with neighboring modes having
highly correlated drift rates (10). Also, as the
comb modes drift across the detector, higher-
order distortions arc evident, which arc the com-
bined result of many effects, such as intrapixel

Fig. 1. Sketch of our experimental setup at the VTT. By superimposing the frequency comb with
light from a celestial body — in this case, the Sun — one can effectively calibrate its emission or
absorption spectrum against an atomic clock. An erbium-doped fiber LFC with 250-MHz mode
spacing (pulse repetition rate) is filtered with a FPC to increase the effective mode spacing,
allowing it to be resolved by the spectrograph. The latter has a resolution of ~0.8 GHz at
wavelengths around 1.5 pm. where our LFC tests were conducted. The LFC was controlled by a
rubidium atomic clock. A continuous-wave (CW) laser at 1583 nm was locked to one comb line and
simultaneously fed to a wavemeter. Even though the wavemeter is orders of magnitude less precise
than the LFC itself, it is sufficiently accurate (better than 250 MHz) to identify the mode number, n.
The FPC length, defining the final free spectral range (FSR), was controlled by feedback from its
output See (10) for further details.

1336

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

REPORTS

Fig. 2. Spectra of the
solar photosphere (back-
ground image) overlaid
bya IK with 15-GHz mode
spacing (white, equally
spaced vertical stripes).
Spectra are dispersed hor-
izontally, whereas the ver-
tical axis is a spatial cross
section of the Sun's pho-
tosphere. The upper pan-
el shows a small section
of the larger portion of
the spectrum below. The
brighter mode labeled
with its absolute frequen-
cy is additionally super-
imposed with a CW laser
used to identify the mode
number (Fig. 1). The fre-
quencies of the other
modes are integer mul-
tiples of 15 GHz higher
(right) and lower (left) in
frequency. Previous cali-
bration methods would
use the atmospheric ab-
sorption lines (dark vertical
bands Labeled "Atm" inter-

leaved with the Fraunhofer

absorption lines), which are comparably few and far between. Also shown in the upper panel is the only
thorium emission line lying in this wavelength range from a typical hollow-cathode calibration lamp.
Recording it required an integration time of 30 min, compared with the LFC exposure time of just 10 ms.
Unlike with the LFC, the thorium calibration method cannot be conducted simultaneously with solar
measurements at the VTT. The nominal horizontal scale is 1.5 x 10 -3 nm pixel -1 with -1000 pixels shown
horizontally in the upper paneL Black horizontal and vertical lines are artifacts of the detector array.

Fig. 3. Precision achieved with our
calibration with a LFC filtered to 3 GHz
On = 12). (A) The position of the trans-
mitted modes, derived from a multi-
Lorentzian fit, plotted against the mode
number. Modes without a corresponding
detector position measurement were deemed
unsuitable for use in calibration because they
fell on large detector artifacts and/or were
overlaid with light from the CW Laser. The
size of one pixel corresponds to 172 MHz at
1583 nm. On this scale, no distortions are
visible. (B) The residuals from a quadratic fit
that gives a RMS residual of 9 m s -1 . The
quadratic fit greatly reduces the residuals
compared to a linear model, whereas higher-
order polynomials do not improve the
performance of the fit significantly (10).
Even with these first LFC recordings from the
VTT, the 9 m s -1 RMS residuals here indicate
better absolute calibration than is achieved
with traditional calibration methods (30).

<ui

-a is

o uo no *o «o too
Mode number (n - 769336)

sensitivity variations. Thus, the VTT data already
show an important advantage of LFC calibration:
The dense grid of high SNR calibration infonna
tion allows the discovery and measurement of com-
plex effects con-elated across the chip and in time.

The first light for frequency combs on astro-
nomical spectrographs has delivered calibration
precision beyond the state of the art. The key
opportunity now is to use LFC spectra to measure
and remove systematic effects in order to reach
the ~1 cm s _I long-term calibration precision,
accuracy, and repeatability required to realize the
redshift drift experiment.

References and Notes

LA.G. Riess ef al. Astron. }. 116. 1009 (1998).

2. S. Perlroutter el oL Astrophys. ). 517. 565 (1999).

3. 0. N. Spergel el ol., Astrophys /. 148 (suppl.). 175 (2003).

4. J. A. Peacock et ol. Nature 410. 169 (2001).

5. 0. J. Eisenstein el ol. Astrophys ). 633. 560 (2005).

6. A. Sandage. Astrophys ). 136. 319 (1962).

7. A. Loeb. Astrophyi. ). 499. Llll (1998).

8. ). Liske et ol. Mon Not. ft. Astron. So c 386, 1192 (2008).

9. L Pasquini et oL in Proceedings of the IAU Symposium.
P. Whitelock, M. Define feld. B. leibundgut. Eds.
(Cambridge Jniv. Press, Cambridge. 2006), voL 232,
pp. 193-197.

10. Materials and methods are available as supporting
material on Science Online.

1L M. Mayor. D. Queloz. Nature 378. 355 (1995).

12. G. W. Marcy. R. P. Butler. Astrophys. ). 464. 1147 (1996).

13. C. Lovis et at.. Nature 441. 305 (2006).

14. C. Lovis ef at., in Proceedings of the SPiE. I. S. McLean,

L Masanori, Eds (2006). vot. 6269. pp. 62690P1-62690P9.

15. J. N. BahcalL E. E. Salpeter. Astrophys ). 142. 1677 (1965).

16. J. < Webb. V. V. Flambaum, C. W. Churchill M. J. Driidcwater.
J. D. Barrow. Phys Rev. Lett 82, 884 (1999).

17. R. I. Thompson. Astron. tett. 16. 3 (1975).

18. M. T. Murphy. ). K. Webb. V. V. Flambaum. Mon. Not ft
Astron. Soc. 345. 609 (2003).

19. H. Chand, R. Srianand. P. Petitjean, B. Aradl, Astron.
Astrophys 417. 853 (2004).

20. S. A. Levshakov ef al, Astron. Astrophys 449. 879 (2006).

21. E. Re inhold et al. Phys Rev. left 96. 151101 (2006).

22. Th. Udem, R. Holmarth. T. W. Hansch. Nature 416, 233
(2002).

23. M. T. Murphy ef at.. Mon. Not. ft Astron. Soc 380. 839
(2007).

24. L H. Schroter. D. Soltau. L Wiehr. Vistas Astron. 28,
519 (1985).

25. T. I. Suer. IEEE). Quantum Electron. 25. 97 (1989).

26. Th. Udem. ). Reichert. R. Holwarth. T. W. Hansch.

Phys. Rev. Lett. 82. 3568 (1999).

27. P. 0. Schmidt. S. Kimeswenger. H. U. Kaeufl. Ptoc. 2007
ESO Instrument Calibration Workshop. ESO Astrophysics
Symposia series (Springer, in press): available at
http://antiv.org/abs/0705.0763.

28. C.-H. Li et at.. Nature 452. 610 (2008).

29. 0. A. Braje, M. S. Kirchner, S. Osteiman, T. Fortier.

S. A. Diddams. fur. Phys ). D 48. 57 (2008).

30. C. Lovis. F. Pepe. Astron. Astrophys. 468. 1115 (2007).

31. We thank ihe Kiepenheuer Institut fur Sonnenphysik staff at
the Vacuum Tower Telescope and the Institute de AstroHska
de Can arias (IAO personnel for their support during the
measurements at the VTT. We especially appreciate the
efforts of M. Colados (IAO and F. Kertaer (ESO). We thank

T. Kippenberg tor CW-laser assistance and ]. liske for advice
on the manuscript. M.T.M. thanks the Australian Research
Council for a QEII Research Fellowship (DP0877998L

Supporting Online Material

www.scie ncema g. org/cgi /conte nt/full/32 1758 94/13 35/DC 1
Materials and Methods
Figs. SI to S4
References

28 May 2008. accepted 25 July 2008
10.112 6/science.ll61030

www.sciencema 9 .or 9 SCIENCE VOL 321 5 SEPTEMBER 2008

1337

REPORTS

Regional Synthesis of Mediterranean
Atmospheric Circulation During
the Last Glacial Maximum

J. Kuhlemann, 1 * E. J. Rohling, 2 I. Krumrei, 1 P. Kubik, 3 S. Ivy-Ochs, 4 M. Kucera 1

Atmospheric circulation leaves few direct traces in the geological record, making reconstructions
of this crucial element of the climate system inherently difficult. We produced a regional
Mediterranean synthesis of paleo-proxy data from the sea surface to alpine altitudes. This provides
a detailed observational context for change in the three-dimensional structure of atmospheric
circulation between the last Glacial Maximum (LGM, -23,000 to 19,000 years ago) and the
present. The synthesis reveals evidence for frequent cold polar air incursions, topographically
channeled into the northwestern Mediterranean. Anomalously steep vertical temperature gradients
in the central Mediterranean imply local convective precipitation. We find the LGM patterns
to be analogous, though amplified, to previously reconstructed phases of enhanced meridional
winter circulation during the Maunder Minimum (the Little Ice Age).

M editerranean climate is determined by
an interplay between atmospheric and
marine processes and strongly differ-
entiated regional topography (/). A wealth ofpaleo-
climate data is available from archives recording
conditions at the sea surface and on land at var-
ious altitudes, making the Mediterranean one of
the few regions in the world where the thermal
and dynamical structure of the lower atmosphere
could be reconstructed for certain past intervals
(2). Such reconstructions are invaluable for vali-
dation of the atmospheric component in climate
models (3). Recent attempts to compare model
simulations with regional proxy data over Europe
during the Last Glacial Maximum (LGM) revealed
substantial disagreement, both among the models
and between models and paleodata (4, 5), high-
lighting the need for model- independent con-
straints on past regional climatic patterns.

The state of the atmosphere in the past is in-
herently difficult to reconstmct. Proxies from ocean
ic sediments record mainly large-scale atmospheric
patterns (6): and terrestrial proxy data, such as
those from peat bogs or lake sediments, can be
biased by local climate, including temperature
inversion and interannual variability (7, 8). The
equilibrium line altitude (ELA) of glaciers con-
tains information on the vertical structure of the
atmosphere, which can be reconstructed by in situ
dating of glacial advances and retreats. Small tem-
perate glaciers in circum-Mcditcrranean mountain
chains are (and were) exposed to well-mixed air
masses and arc known to have been sensitive to
even small changes of the ELA, typically respond-
ing by advancing or retreating within periods tanging
from several years to decades (9, 10).

'institute for Geosciences Univenityof Tuebingen, Sigivartstrasse
10, D-72076 Tuebingen, Germany. ^School of Ocean and Earth
Science, National Oceanography Centre, Southampton $014
3ZH, UK. ’institute of Particle Physics, HPK H30, ETH Zurich,
CH-8093 Zurich, Switzerland, institute of Particle Physics, HPK
HZ7, ETH Zurich, CH-8093 Zurich, Switzerland.

•To whom correspondence should be addressed. E-mail:
kuhlemann@uni-tuebingen.de

The ELA responds to both temperature and
precipitation change (9, 10), and it is possible to
differentiate between these two factors only in
particularly well-studied regions, such as Corsica
(data supplement SI and figs. S2 and S3). For
Corsica, we present new information on the LGM
ELA, including a deconvolution of the two main
controlling processes (fig. S4, B and C). For the
ELA depression of LGM glaciers in the wider
Mediterranean region, we used previously pub-
lished information (table S2), which, as a first-
order end-member solution, we calculated as pure
temperature change, using a standard free atmo-
spheric lapse rate of a 6.5°C decrease per kilometer
(6.5°C/km) of increasing elevation. The potential
overprint of precipitation changes was then con-
sidered where anomalous results were found. Tire
error ranges on the resultant ELA reconstructions

(Fig. I) amount to up to ±100 m in Corsica and
± 150 m in other Mediterranean mountains (fig. SI
and Fig. 2). We thus developed a regional synthe-
sis of glacial vertical temperature gradients in the
lower atmosphere. Paleo flora-based temperature
reconstructions for a variety of terrestrial sites at
lower altitudes around the Mediterranean (7, 8)
(Fig. 2 and fig. S3) were used to validate and com-
plement our ELA -based temperature reductions
and precipitation patterns.

Next, we compared the ELA-based LGM cool-
ing at alpine altitudes with estimates of LGM reduc-
tion of Mediterranean sea surface temperatures
(SSTs) derived from the difference between long-
term instrumental averages (//) and glacial SST
reconstructions based on fotaminiferal assemblages
(12, 13) and alkenone data (14) (Figs. 1 and 2).
Such direct comparison between SST and ELA
changes is warranted for the Mediterranean basin,
where SSTs generally are closely related to air tem-
perature and the insolationfradiation balance (IS).

The combination of data on LGM cooling at
sea level (SST proxies) and higher altitudes (ELA
depression) provides direct constraints on the
vertical structure of the LGM atmosphere. When
comparing the temperature equivalent of the ELA
depression with SST reduction in the LGM rel-
ative to the present (Fig. 2), we consider that a
shift of similar magnitude would indicate a con-
stant atmospheric lapse rate. Stronger relative
reduction of SST would imply a lapse rate of less
than 6.5°C/km, supporting more stable atmospheric
stratification. A lesser relative SST reduction would
imply a lapse rate steeper than 6.5°C/km, poten-
tially enhancing the instability of the atmosphere,
driving convection and consequent precipitation.

Our analysis (figs. SI and S2) reveals an
LGM pattern of southw ard-extending lobes of
ELA depression in mountainous regions of Italy

Fig. L Map of the ELA in the western-central Mediterranean region during the phase of maximum
glacier expansion during the LGM (probably at ~23,000 years before the present) and of average
annual SSTs and ELA during the LGM. The error range of the ELA estimate is ±150 m for the
Mediterranean in general and ±100 m in Corsica.

1338

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

REPORTS

and the Dinarides, which suggests frequent higher-
altitude southward advances of polar air (Fig. 1).
Iberia is characterized by a steep gradient from
the northern and northwestern coastlines toward
the interior and southeast, which probably results
predominantly from barrier effects of near-coastal
mountain ranges (Fig. 1). The data from Corsica
especially identify a lobe of ELA depression that
extends over the Gulf of Lions toward the south
and east (Fig. 1), indicating a substantial invasion
of polar air from the north. The temperature dif-
ference inferred from the recent ELA ( 76) and
our LGM reconstruction (fig. S3) generally de-
creases from north (10° to 1 1°C) to south (6° to
7°C) (Fig. 2), in agreement with previous re-
constructions of a steeper glacial meridional temper-
ature gradient (76, 17). The temperature differences
calculated from the glacial ELA depression, relative
to the present, generally agree with lower-altitude
temperature reconstructions from palcofloral data
(7, 8) (Fig. 2).

The present-day SST distribution and surface
circulation in the western Mediterranean basin
arc strongly affected by northwesterly winds, par-
ticularly in the Gulf of Lions (75). As a conse-
quence, cool waters are frequently upwelling in
the Gulf of Lions (77, IS). Surface currents arc
deflected by coastlines, and their strength and
flow direction vary seasonally in response to sur-
face winds and the superimposed atmospheric cir-
culation (77, 75). Glacial SST values calculated from
foraminiferal assemblages (12, 13) and alkcnonc
data (14) display a roughly similar distribution to
that of modem SSTs, albeit with a stronger west
east gradient due to stronger cooling in the north-
western Mediterranean than in the central and

eastern parts of the basin (12) (Fig. I and data sup
plement S3). The extraordinary cooling centered
on the Gulf of Lions suggests frequent and/or
more persistent northerly incursions of cold polar
air, probably channeled through the Rhone valley
at low elevation (14, 18), and between the glaciated
Alps and the Pyrenees at higher elevation, as sug-
gested by our ELA reconstructions.

Figure 2 compares the spatial pattern of the
LGM reduction of SST (relative to the present) with
that of atmospheric temperature as derived from our
ELA reconstruction. This reveals that both SST
and ELA determincd atmospheric temperatures
(Tela) underwent similar (within ±2°Q changes,
relative to the present, across the northern Bay of
Biscay and the western sector of the western Medi-
terranean. LGM SST seems less reduced than Tela
in the Atlantic Ocean offshore of Iberia and Mo-
rocco, which probably reflects the southward dis-
placement of the relatively warm Gulf Stream during
glacial times (3-6, 13, 19). In the central and (to a
lesser extent) eastern Mediterranean, glacial SST
appears to have dropped considerably less than
Tela (Fig. 2). The notable warm anomaly in the
central basin can hardly be attributed to the advec-
tion of warm surface waters from the western
basin because of land barriers. In fact, a notable
cool SSTanomaly is seen to the southeast of Sardinia,
which may reflect leeward upwelling triggered
by northwesterly winds (Fig. 1). We propose that
the advcction of warm desert air from the Sahara
and relatively cloud-free subtropical conditions
over the central/eastem basin largely account for
the minor LGM cooling of SSTs in this region.
The fret that glacial SST dropped considerably
less than calculated Tela over part of the Mediter-

ranean suggests that the atmospheric lapse rate had
noticeably steepened: up to ~10°C/km north of
Corsica, ~9°C/km in tire southern Adriatic Sea,
and ~8.5°Okm in the central Mediterranean basin.
Given that we applied an initial end-member ELA
transformation to (only) temperature changes, using
a standard lapse rate of 6.5°C/km, it is clear that
increased convective precipitation must be inferred
to explain the noticeably sleeper rates diagnosed
in these specific regions.

The spatial distributions of SST, Tela, and of
the SST- Tela difference in the western -central
Mediterranean during the LGM arc found to be
roughly similar to those in the present, although
meridional gradients were enhanced during the
LGM (Figs. 1 and 2). Hence, it is not unreasonable
to expect that cyclones followed similar prefer-
ential storm tracks across the basin as well, which
contrasts with previous suggestions of northeast
directed cyclone tracks from the Alboran Sea toward
the southern flank of the Alps (20). During cold
periods such as the LGM, cold northerly air out
breaks over the western basin were probably more
frequent (12, 17, 18). The pronounced southward
cold (polar air) expansion toward northwest Africa
(Figs. 1 and 2) would have triggered cyclogenesis
over the relatively warn Mediterranean waters,
causing flows of desert air toward the north and
northeast, as indicated by the no rth ext ending lobe
of the ELA in southeastern Europe (Fig. 1). This
would be consistent with observations of enhanced
wind-blown dust supply from the Sahara into the
eastern Mediterranean during glacial times (27).

Even though we compare glacial conditions
(the LGM) with interglacial conditions (the
present), we observe that the reconstructed prop-
erty distribution patterns, particularly the prefer-
ential flow of polar air masses, are pervasive through
time (Figs. 1 and 2). Indeed, these features appear
to be strongly fixed by the land/sea distribution
and topography, which are virtually invariant on
the time scales considered. Outbreaks of polar air
masses over the western Mediterranean are typ-
ically founded between the Alps and the Pyrenees,
both at present and during the LGM, causing con-
ditions conducive to cyclogenesis over the Gulf
of Genoa. The tunneling effect may have been
stronger with glaciated mountains, as the ice rose
several hundreds of meters above the lower water-
sheds (20), and Arctic air masses would also have
invaded the western Mediterranean more frequent
ly and/or persistently than today, because of the
more southerly position of the polar front during
the LGM (3-5, 19, 22). The incursion of cold air
masses would have favored the convection of
moist air; especially in regions with relatively
warm (less reduced) SST, so that we would pre
diet considerable local LGM precipitation in
Corsica, the Apennines, the Dinarides, and Greece,
especially at the upwind flank of mountain ranges
and close to the coast This would be a suitable
mechanism to explain steeper horizontal precipi-
tation gradients during the LGM relative to foe
present which indeed are suggested by our data
for the steep mountainous margins of northern

Fig. 2. Map of the temperature difference between recent and LGM SSTs (in black) and the temperature
equivalent of the ELA depression (6.5°C/km lapse rate; in blue), respectively. The error range of this estimate
is ±1°C for the Mediterranean in general and ±0.7°C in Corsica. In orange-colored marine regions, LGM SSTs
were lowered significantly less than temperatures in the mid-troposphere, relative to the present This
implies an anomalously steep lapse rate and unstable layering of the lower troposphere. Atmospheric
cooling values for tow-elevation terrestrial sites based on paleoftoral estimates are given for comparison.
Small symbols indicate larger error and large symbols lesser error of the temperature estimate. A, change in.

www.sciencema 9 .org SCIENCE VOL 321 5 SEPTEMBER 2008

1339

REPORTS

Corsica (tabic SI and fig. SI). This island’s diy
northern interior today receives -30% less pre-
cipitation than its margins (fig. S2A), whereas this
difference was -50% during the LGM (fig. S4Q.
Although this prediction cannot (yet) be con-
firmed with the data available outside Corsica, it
does agree with patterns seen in LGM reconstruc-
tions with the high-rcsolution climate model HadRM
( 23 ) As mentioned above, locally enhanced pre-
cipitation would largely reduce the local lapse
rate, so that much of the initially (first-order) in-
ferred temperature anomaly pattern in fact re-
flects the impact of precipitation anomalies.

Although care must be taken not to simply
ascribe past regional property distributions to mod-
em climate oscillation patterns ( 24 ), it remains
useful to consider instrumental records and proxy
data in order to develop a sense of realistic anal-
ogous climate patterns over the study region (25).
The contrast between strongly reduced SST in
the western basin and much less reduced SST in
the central Mediterranean basin during the LGM
(Fig. 1 ) indicates a preferentially meridional geo
strophic circulation, with a polar trough that fre-
quently protruded into the western Mediterranean.
Such a circulation is favored by northward exten-
sion of the Azores High toward Iceland (North
Atlantic ridge) or Greenland, blocking moisture
supply by the westerlies. It is further enhanced by
expansion and intensification of the Siberian High
in winter during glacial times ( 26 ). A similar
configuration is thought to have been common
during the late Little Ice Age, notably the Maunder
Minimum (2, 27). The invasion of polar air as
shown by our data, channeled by the topography
of mountain ranges and ice sheets in Europe,

would have generated cyclone formation in the
Gulf of Genoa more frequently than at present,
enhancing precipitation along various storm tracks
in easterly directions. Our observations do not sup
port a straightforward zonal LGM atmospheric cir-
culation, as inferred from climate models ( 19 , 28 ).
Instead, we propose that frequent meridional cir-
culation during cold seasons (characterized by the
LGM ELA pattern) may have alternated with more
zonal circulation during warm seasons. A mote com-
prehensive quantitative assessment of the prcfcren
tial LGM atmospheric circulation requires die use
of both nested model simulation and high-rcsolution
global climate model studies ( 4 , 5 , 8 , 28 ), which
should fully resolve the changing topography of
glaciated mountain ranges and ice sheets. The vali-
dation of such models with our three-dimensional
LGM climate proxy data ranging from the sea sur-
face to alpine altitudes is a great future challenge.

References and Notes

L E. Xoplaki. ). F. Gonrater Rouco, J. luterbacher.

H. Wanner Cbm. Dyn. 23. 63 (2004).

2. J. Luterbacher el at., in The Mediterranean Gimate: An
Overview of the Main Characteristics and Issues. P. Lionelto.
P. Malanotte-Rirroti, R, Bosoolo, Eds. (Elsevier, Amsterdam,
2006). pp. 27-148.

3. C0HMAP Members. Science 241. 1043 (1988).

4. M. Kageyama el at.. Quat Sci. Rev. 23. 2082 (2006).

5. G. Ramstein et at.. Clim. Past 3. 331 (2007).

6. G. Bond et at.. Nature 365. 143 (1993).

7. 0. Peyton et at.. Qual Res. 49. 183 (1998).

8. H. Wu, J. Guiot S. Brewer. Z. Guo. Ctim. Dyn. 29. 211
(2007).

9. H. Kerschner. G. Kaser. R. Sailer. Ann. Godot 31, 80 (2000).

10. D. Steiner etal.. Cbm. Change. 10.1007/S10584-008-9393-1
(2008).

11. M. Conkright et at.. World Ocean Atlas 1998 CD ROM
Data Set Documentation (Technical Report 15. NODC
internal Report Silver Spring. MD. 1998).

12. A. Hayes, M. Kucera. N. Kallel L Sbatti. E. J. Robi ng.
Quat So. Rev. 24. 999 (2005).

13. A Paul C Schafer-Neth, Patcieoceanography 18. 1058 (2003).

14. L Cacho. J. 0. Grimalt M. Canals. ). Mar. Syst. 33 34.
253 (2002).

15. N. Pinardi, E. Masetti. Pataeogeogr. Pataeoctimatol
PataeoecoL 158. 153 (2000).

16. B. Messerli, Geogr. Hetv. 22. 105 (1967).

17. V. Masson Detmotte et at.. Ctim. Dyn. 26. 513 (2006).

18. E. J. Rohling. A Hayes, 0. Kroon, S. Oe Rijlc.

W. J. Zachariasse. Palaeoceanograpby 13. 316 (1998).

19. M. Kageyama. F. D. Andrea. G. Ramstein. P. J. Valdes,

R. Vautard. Ctim. Dyn. 15. 773 (1999).

2tt a Florineth, C. Schluchter, Quat. Res 54. 295 (2000).
21 ]. C. Larrasoana. A P. Roberts, E. J. Rohling. M. Winkthofer,
R. Wehausen. Cbm. Dyn. 21. 689 (2003).

22. U. Pitaumann et at.. Pataeoceanography 18. 10.1029/
2002PA000774 (2003).

23. A lost et at.. Clim. Dyn. 24. 577 (2005).

24. F. Justino. W. R. Pettier. Geophys Res Lett. 32. iai029/
2005G1023822 (2005).

25. C. Cassou. L Temay. J. W. Hurrell. C. Deser, J. Clim. 17.
1055 (2004).

26. E. ]. Rohling. P. A Mayewski. P. Challenor. Clim. Dyn. 20.
257 (2003).

27. J. Jacobeit. P. Jons son, l. Barring, C. Beck. M. Ekstrom.
Oim. Change 48, 219 (2001).

28. A Laine' et at.. Clim. Dyn.. 10.1007/s00382 008 0391-9
(2008).

29. We gratefully acknowledge funding by the German
Science Foundation (DFG project KU 1298/7) and
the UK Natural Environment Research Council's
thematic program Quantifying the Earth System (QUEST).

Supporting Online Material

www.scie ncema g.orgfcgi/conte nt/f utl/11 576 38/DCl

Data Supplements SI to S3

Figs. SI to S 7

Tables SI and S2

References

11 Match 2008: accepted 21 July 2008
Published online 31 July 2008;

10.112 6/science.ll57638

Include this information when citing this paper.

Kinematic Constraints on Glacier
Contributions to 21st-Century
Sea-Level Rise

On the basis of climate modeling and analogies with past conditions, the potential for multimeter
increases in sea level by the end of the 21st century has been proposed. We consider glaciological
conditions required for large sea-level rise to occur by 2100 and conclude that increases in excess
of 2 meters are physically untenable. We find that a total sea-level rise of about 2 meters by 2100
could occur under physically possible glaciological conditions but only if all variables are quickly
accelerated to extremely high limits. More plausible but still accelerated conditions lead to total
sea-level rise by 2100 of about 0.8 meter. These roughly constrained scenarios provide a "most
likely" starting point for refinements in sea-level forecasts that include ice flow dynamics.

W. T. Pfeffer, 1 * J. T. Harper, 2 S. O'Neel 3

E ustatic land ice contributions to sea-level
change come from surface mass balance
(SMB) losses and discharge of ice into the
ocean through marine-terminating glaciers. Dy
namicaliy forced discharge, via fast flow and calv-
ing of marine-terminating glaciers allowing rapid
land-to occan transfer of ice, is well known from
studies of temperate marine-terminating glaciers

( 1 - 4 ) and is observed in Greenland (5-7). The
consensus estimate of sea-level rise (SLR) by
2100 (0.18 to 0.6 m) that was published in
the Intergovernmental Panel on Climate Change
(IPCQ Fourth Assessment (5) excluded dynamic
effects on the grounds that present understanding
of the relevant processes is too limited for reliable
model estimates. Because modeling ( 9 ) and paleo-

climate comparisons ( 10 ) have yielded multimeter
per century estimates of SLR, similar increases
have been inferred as a viable 21st-century sce-
nario. Also argued is that feedbacks unaccounted
for in the IPCC estimate could quickly cause sev-
eral meters of very rapid SLR ( 11 , 12 ).

Accurate SLR forecasts on the century time
scale are imperative for planning constructive
and cost-effective responses. Underestimates will
prompt inadequate preparation for change, whereas
overestimates will exhaust and redirect resources
inappropriately. Raising California Central Valley
levees only 0.15 m, for example, will cost over
$1 billion ( 13 ); the nonlincarly increasing costs
of raising levees 2 m or more without clear and
compelling cause would entail enormous expend-
itures otherwise used for different responses as
demanded by a smaller but still significant SLR.

We address the plausibility of very rapid SLR
from land ice occurring this century. We give

'institute of Arctic and Alpine Research, University of
Colorado, Boulder, CO 80309, USA. department of
Geosciences, University of Montana, Missoula, MI 59812,
USA. 3 Scripps Institution of Oceanography, University of
California San Diego, San Diego, CA 92093, USA
•To whom correspondence should be addressed. E-mail:
pfeffer@tintin.colorado.edu

1340

5 SEPTEMBER 2008 VOL 321 SCIENCE wvwv.sciencemag.org

REPORTS

particular emphasis to Greenland because of its
vulnerability to ongoing Arctic wanning and
meltwater-related feedbacks, recent accelerations
of ice motion, and its large volume reductions
during the last interglacial (14). By using a simple
kinematic approach, we determined Greenland
and Antarctic outlet glacier velocities required to
achieve various magnitudes of SLR by 2100.

To begin, we postulated sea- level increases of
2 and 5 m by 2100 forced solely by Greenland.
The total water mass losses required to achieve
these targets are 7.24 x 10 5 Gt and 1 .81 x 1 0 6 Gt,
respectively (Table 1). Of this mass, 2.58 x lO -1
Gt (less than 4% of the total for 2 m SLR) will be
lost as SMB by 2100, which we estimated by
integrating present-day values of mass loss for-
ward at present-day rates of change (15), with
present-day SMB estimated at 30% of total present
day rate of mass loss (5). Because future SMB is
highly uncertain, we also scaled total SMB losses
up by a factor of 10 to investigate the effect of
uncertainty in SMB. Adjusting total mass losses
for SMB contributions yields the mass to be dis-
charged through marine-based outlets (Table 1).
Even when scaled up by an order of magnitude,
SMB is a very small traction of the total loss
required to produce the targeted SLR. Thus, even
large uncertainties in future SMB have little in-
fluence on this calculation.

Rapid, dynamically unstable discharge of ice
through calving is restricted to glaciers with beds
based below sea level. We identified and cal-
culated tiie aggregate cross-sectional area of Green-
land's marine-terminating outlet glaciers by using
surface and bed topography (16) and measured
ice velocities (5) to identify all potential pathways
for rapid discharge, including channels presently
flowing rapidly as well as potentially unstable
channels (Fig. I and table SI). Cross-sectional
areas (gates) for each outlet were calculated at the
point of greatest lateral constriction by bedrock in
the glacier's marine-based reach. Ice stream widths
in Antarctica can vary in time, but for Greenland
outlet glaciers cross-sectional areas are constrained
almost entirely by bedrock topography. Of the
290 km 2 total aggregate gate cross-sectional area,
we identified 170 km 2 as the aggregate marine-
based gate area where drainage to the ocean is not
blocked by near-coastal sills standing above present
day sea level. All dynamic discharge (Table 2)
must pass through these gates by 2100 to meet 2-
to 5 m SLR targets. We considered four scenarios:
velocities were calculated for both the "marine
based” gate (170 km 2 ) and the "total aggregate”
gate (290 km 2 ) given both projected SMB and
10>: inflated SMB losses. We then considered
whether those velocities arc realistic.

Coarseness of the digital elevation models
(DEMs) used for surface and bed topography (16)
led to uncertainties in the calculated gate areas,
which may be substantial but cannot be evaluated
directly. We accounted for a potential underestimate
of gate area with a calculation using the total
aggregate gate area. The total aggregate area
exceeds the more relevant marine- based gate area

by 70%. Uncertainties arising from the DEM more
likely conceal small unresolved channels than large
ones, so the actual gate areas may be smaller than
we calculate (thus yielding higher velocities).

The present-day average velocity of all Green-
land outlet glaciers is 0.56 kmfyear when weighted

by drainage basin area or 1.23 km/year when
weighted by gate cross-sectional area. The two
weighted averages are different because gate cross-
sectional area does not scale with drainage basin
area. Average (present day to 2 1 00) outlet glacier
speeds required to meet 2- and 5-m SLR targets

Table 1. Fluxes and discharges for SLR targets. Q indicates total discharge to 2100 (Gt); q, 0
converted to ice flux rate, total to 2100 (km 3 /year); Q 1( total dynamic discharge less SMB to 2100
(Gt); q x , 0 1 converted to ice flux rate, total dynamic flux less SMB to 2100 (km 3 /year); Q 2 , total
dynamic discharge less lOx SMB to 2100 (Gt); and q 2 , 0? converted to ice flux rate, total dynamic
flux less lOx SMB to 2100 (km 3 /year).

SLR target SLR mm/year Q 9 Pi 9i 0 ? q?

2 m 21.5 724,000 8,650 698,164 8,341 652,464 7,795

5 m 53.8 1,810,000 21,625 1,784,165 21,317 1,738,464 20,770

90°W 60°W 20°W 0° 10°E

Fig. 1. Map showing Greenland and outlet glacier gates; marine-based gates are shown as dark
green and nonmarine as black. Regions below sea level are colored blue, ice velocities at -2000 m
elevation from (21) shown by red dots.

Table 2. Required velocities of Greenland gates for SLR targets.

SLR target

2 m
5 m

Present-day
marine-based
gates (km/year)
lx SMB

49

125

All present-day
discharge gates
(km/year)
lx SMB

28.7

73.4

Present-day
marine-based
gates (km/year)
lOx SMB

45.8

122

All present-day
discharge gates
(kmfyear)
lOx SMB
26.8
7L5

www.sciencema 9 .or 9 SCIENCE VOL 321 5 SEPTEMBER 2008

1341

REPORTS

range from 26.8 km/ycar to 125 km/year, de-
pending on the scenario considered [Table 2 and
supporting online material (SOM)]. These veloc-
ities must be achieved immediately on all outlets
considered and held at that level until 2100.
Delays in the onset of rapid motion increase the
required velocity further (fig. SI).

The scenario velocities far exceed the fastest
motion exhibited by any Greenland outlet glacier.
For example, the near-doubling of ice discharge
from Jakobshavn Glacier in 2004-2005 was
associated with an acceleration to 12.6 km/year
(7). Similarly, a temporary 80% increase in the
speed near the terminus of Kangerdlugssuaq
produced a velocity of 14.6 km/year (5). A com-
parison of calculated (Table 2) and observed
(1 .23 km/year) average velocities shows that cal-
culated values for a 2-m SLR exceed observa-
tions by a factor of 22 when considering all gates
and inflated SMB and by a factor of 40 for the
marine gates without inflated SMB, which we
consider to be the more likely scenario. With the
exception of discharge through all gates at in-
flated SMB (26.8 km/year), none of the velocity
magnitudes shown in Table 2 has ever been
observed anywhere, even over short time periods.
The highest observed velocities have occurred at
surging glaciers, including circa (ca.) 70 m/day
(25.5 km/year) at Variegated Glacier (1 7) and 1 05
m/day (38.3 km/year) at Medvezhiy Glacier (IS),
but were held only for brief periods (hours to
days). Although no physical proof is offered that
the velocities given in Table 2 cannot be reached
or maintained over century time scales, such be-
havior lies far beyond the range of observations
and at the least should not be adopted as a central
working hypothesis.

Calculations are made only for Greenland
because Greenland’s outlet glaciers are well
constrained by bed topography, which (despite the

Table 3. SIR projections based on kinematic sce-
narios. Thermal expansion numbers are from (22).

SLR equivalent (mm)
Low 1 Low 2 High 1

Greenland

Dynamics

93

93

467

SMB

71

71

71

Greenland total

16S

165

538

Antarctica

PIG/Thwaites dynamics

108

394

Lambert/Amery dynamics

16

158

Antarctic Peninsula

12

59

dynamics

SMB

10

10

Antarctica total

146

128

619

Glaciers/ice

caps

Dynamics

94

471

SMB

80

80

GIC total

174

240

551

Thermal expansion

300

300

300

Total SLR to 2100

785

833

2008

uncertainties mentioned) is well known in com-
parison to much of the Antarctic Ice Sheet and
the Antarctic Peninsula and virtually all of the
marine-terminating glaciers and ice caps (GIQ
exclusive of Greenland and Antarctica. In order
to estimate how these constraints influence pro-
jections of total SLR to 2100, we calculated a
zero-order range of custatic SLR from all land ice
sources. Because marine-grounded channels arc
not well defined in many other locations, we made
approximations and scaling arguments to arrive
at a range of values for total eustatic SLR and,
including reasonable projections of steric SLR, a
range of estimates of total SLR to 2100.

Most of the marine-based ice in West Ant
arctica is held behind the Ross and Filchner-
Ronne ice shelves, which we consider unlikely to
be removed by climate or oceanographic pro-
cesses within the next century [e.g., (19)]. The
Amundson Coast basin [including Pine Island
Glacier (PIG) and Thwaites Glacier], however, is
not confined by large ice shelves and contains
about 1.5 m sea level equivalent (5.43 * 10 s Gt)
(20). The aggregate cross-sectional gate area of
PIG and Thwaites Glacier is ca. 120 km 2 (20).
The average velocity in this region is 2 kra/year
(table S2), higher than the average velocity of all
Antarctic ice streams [0.65 km/year (19)\ An
average (present day to 2100) velocity of 53.6
km/year is required to discharge 1 .5 m sea-level
equivalent through the PIG and Thwaites glacier
gates by 2100, again far greater than any ob-
served glacier velocity.

We present three scenarios by combining likely
projection methods that we believe roughly bracket
the range of potential near-future SLR outcomes
(SOM). These are not true limiting cases but give a
good sense of the potential variability of total SLR
due to dynamic discharge effects.

SLR scenario Low 1 represents a low-range
estimate based on specific adjustments to dy-
namic discharge in certain potentially vulnerable
areas. We assumed a doubling of outlet glacier
velocities in Greenland and PIG/Thwaites within
the first decade and no change from present-day
discharge values at Lambert/Amery. SMB for
Greenland, the Antarctic Peninsula, and GIC was
accelerated at present-day rates of SMB change,
and, lacking more directly applicable constraints,
dynamic discharge for the Antarctic Peninsula
and GIC was calculated by scaling dynamic dis-
charge to SMB by using the ratio of 1.31 as
computed for Greenland (SOM). The net result,
including thermal expansion, is 785 mm by 2100
(Table 3).

A second low-range scenario (Low 2) shows
the effect of varying our assumptions; for this, we
simply integrated presently observed rates of
change forward in time We calculated Greenland’s
contribution as for Low I but accelerated foe
present day net discharge for Antarctica (East/West/
Antarctic Peninsula) forward at foe present-day rate
of change given by (19). The GIC contribution was
also calculated by accelerating the present-day net
discharge at the current rate of change, with values

from (15). The net result, including thermal expan-
sion, is 833 mm by 2100 (Table 3).

SLR scenario High 1 combines all eustatic
sources taken at high but reasonable values. No
firm highest possible value can be determined
for SMB or dynamics; the values chosen repre-
sent judged upper limits of likely behavior on the
century time scale. Greenland SMB was acccl
crated at present-day rates of change, but dy-
namic discharge was calculated by accelerating
outlet glacier velocities by an order of magnitude
in the first decade. In Antarctica, PIG/Thwaites
was accelerated from present-day net discharge
(19) in the first decade and held thereafier to the
highest outlet glacier velocity observed anywhere
[14.6 km/year (6)], and Lambert/Amery was ac-
celerated from present-day net discharge (19) in
the first decade by an order of magnitude and
held thereafter. Antarctic Peninsula and GIC
were calculated by scaling dynamic discharge at
the dynamics- to- SMB ratio computed for Green-
land; this ratio is larger (6.42) that in case Low 1
because Greenland’s dynamic discharge is larger.
The net result, including thermal expansion, is
2008 mm by 2100 (Table 3).

On the basis of calculations presented here,
we suggest that an improved estimate of the range
of SLR to 2100 including increased ice dynamics
lies between 0,8 and 2.0 m. We emphasize that
assumptions made to arrive here contain sub-
stantial uncertainties, and many other scenarios
and combinations of contributions could be con-
sidered. However, the net eustatic SLR from other
combinations explored fell within foe range given
in Table 3. Hence, these values give a context and
starting point for refinements in SLR forecasts on
the basis of clearly defined assumptions and offer
a more plausible range of estimates than those
neglecting the dominant ice dynamics term. Cer-
tain potentially significant sinks and sources of
SLR, such as terrestrial water storage, are still
absent altogether. Among the uncertainties ex
plored, the potential for dynamic response from
GIC is comparable in magnitude to dynamic re-
sponse from Greenland or Antarctica but is excep-
tionally poorly constrained by basic observations.
Without better knowledge of the number, size,
and catchment areas of marine-based outlet gla-
ciers in the GIC category, improvements on the
estimates made here will be very difficult.

References and Notes

1 S. O' Neel. W. T. Wetter. Geophys. ties. Lett. 34. 122502
(2007).

2. M. F. Meier. A. Post ). Geophys. Res. 92. 9051 (1987).

3. M. Meier et at.. ). Geophys. tte i 99. 15219 (1994).

4. B. Kamb et at.. ). Geophys. ties. 99. 15231 (1994).

5. L ftignot P. Kanagaratnam. Science 311. 988 (2008).

6. i. M. Hemal, I. Joughtn. T. A. Scambos. Science 315. 1559
(2007); published online 7 February 2007 (10.1128/
science. 11 38 4 7 8).

7. I Joughin, W. Abdalati. M. Fahnestock. Nature 432. 808
(2004).

8. IPCC. Climate Change 2007: The Physical Science Basis.
Contribution of Working Group I to the Fourth Assessment
Report of the Intergovernmental Panel on Climate
Change. S. Solomon et at.. Eds. (Cambridge Jniv. Press.
Cambridge. 2007).

1342

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

REPORTS

9. Bw l. Otto Bliesner et at.. Science 311. 1751 (2006).

10. J. T. Overpeck el at., Science 311. 1747 (2006).

11. j. Hansen. Environ. Res. tett 2. 024002 (2007).

12. J. Hansen el ol.. Rhilos. Trans. R. Soc. London Set. A 365.
1925 (2007).

13. J. Mount. R. Twiss, San Francisco Estuary Watershed Sci.
3. 1 (2005).

14. L Bard ef at.. Nature 382. 241 (1996).

15. M. F. Meier el at.. Science 317. 1064 (2007); published
online 18 July 2007 (10.112 6/science.U41906).

16. J. L Bamber, R. L. Layberry. S. Gogineni , ). Geophys. Res.
106. 33773 (2001).

17. B. Karrb et at.. Science 227. 469 (1985).

18. E. R. Hope. "English translation T409R." Defense
Research Board. Ottawa. Canada. 1963 [translation
from L D. Dolgushin, S. A. Yevteyev. A. N. Krenke.

K. P. Rototayev, M. M. Svatkov, Rriroda 11. 84 (1963)|.

19. E. R gnot et at.. Nat. Geosci 1. 106 (2008).

20. E. Rignot. personal communication (2007).

21. R. Thomas et at.. Science 289. 426 (2000).

22. G. A. Meehl et oL in (8).

23. This work was supported by NSF grants OPP-0827345 and
OPP-06223S1 and a University of Colorado Faculty
Fellowship (W.T.P.), NSF grants OPP-G612S06 and

0PP-0454789 (|.T.H.>. and the Greens Foundation at Scripps
institution of Oceanography (S.O.). E. Rignot generously
supplied Greenland surface velocity data.

Supporting Online Material

www.sciencemag.org/cgi/content/full/32175894/1340/DCl
Materials and Methods
Fig. SI

Tables SI and S2
References

14 April 2008: accepted 18 July 2008
10.112 6/sdence.ll59099

Apobec3 Encodes Rfv3, a Gene
Influencing Neutralizing Antibody
Control of Retrovirus Infection

Mario L. Santiago, 1 Mauricio Montano, 1 * Robert Benitez, 1 * Ronald J. Messer, 2 Wes Yonemoto, 1
Bruce Chesebro, 2 Kim J. Hasenkrug, 2 t Warner C. Greene 1 ' M t

Recovery from Friend virus 3 (Rfv3) is a single autosomal gene encoding a resistance trait that
influences retroviral neutralizing antibody responses and viremia. Despite extensive research for 30
years, the molecular identity of Rfv3 has remained elusive. Here, we demonstrate that Rfv3 is
encoded by Apobec3. Apobec3 maps to the same chromosome region as Rfv3 and has broad
inhibitory activity against retroviruses, including HIV. Not only did genetic inactivation of Apobec3
convert R/v3-resistant mice to a susceptible phenotype, but ApobecJ was also found to be naturally
disabled by aberrant messenger RNA splicing in /?/v3-susceptible strains. The link between Apobec3
and neutralizing antibody responses highlights an Apobec3-dependent mechanism of host
protection that might extend to HIV and other human retroviral infections.

T he study of viral resistance factors has
provided important insights into the evo-
lutionary strategies of defense used by
mammalian hosts (7-5). Recovery from Friend
vims (FV) gene 3 (Rfv3) was first identified as a
resistance trait in 1978 (6, 7), and later genetic
studies showed that the phenotypes of decreased
viremia and FV-specific neutralizing antibody
responses segregated as a single gene (8).
Because the generation of neutralizing antibodies
is critical for recovery from FV infection (7, 9)
and a desired but often unrealized outcome in
various retroviral infections, including HTV-1, we

have focused our efforts on identifying the gene
encoding R/v3. The Rfv3 locus maps to a 0.83-
centimoigan region of chromosome 1 5 (fig. S 1 A)
(10-12), which contains at least 61 annotated
genes (table S 1), one of which is murine Apobec3
(mA3), a member of a family of dcoxycytidine
deaminases with antiretroviral and antiretroelc-
ment activity [as reviewed in (73)]. This fact,
along with the presence of substantial polymor-
phism in mA3 (table SI), led us to consider mA3
as a prime candidate for RJv3.

Because Rfv3 has no described in vitro phe-
notype, ora investigation required the generation

of m/13-deficient mice (14). First, an inactivated
mA3 gene (fig. SIB) was introduced into the
Rfv3^ C57BL/6 (B6) background to test its
ability to act as a defective Rjv3 allele in matings
with Rfv^ mice (Table 1). Because the Rfv3
resistance trait is dominant over susceptibility (7),
Rjv3 H * Fi offspring should control viremia and
mount effective neutralizing antibody responses.
Conversely, if mA3 encodes Rfv3, then the gene
from a mA3~^~ parent will be null, and the
resultant F j offspring with an RfvJ* genotype are
predicted to exhibit higher levels of viremia and
weaker neutralizing antibody responses. To test
this possibility, B6 x BALB/c Fj offspring were
infected with FV and plasma viremia levels were
measured. At 7 days post infection (dpi), the Fj
mice containing an inactivated mA3 gene ex-
hibited levels of viremia 1 5 times as high as their
congcnic partners carrying the wild-type mA3
allele (Fig. 1 A). These high viral loads in mAf Fj
mice were comparable to FV levels found in fully
susceptible Rjv3 ** BALB/c parental mice. Thus,
mA3 is a restriction factor contributing to the

'Gladstone Institute of Virology and immunology. San
Francisco, CA 94158, USA. laboratory of Persistent Viral
Diseases. Rocky Mountain Laboratories, National Institute
of Allergy and Infectious Diseases, Hamilton, MT 59840,
USA. 'Department of Medicine, University of California,
San Francisco, CA 94143-1230, USA. “Department of Micro
biology and Immunology, University of California, San Frandsco,
CA 94143-1230, USA.

•These authors contributed equally to this work.
tTo whom correspondence should be addressed. E-mail:
khasenkrug@nih.gov (K.J.H.); wgreene@gladstone.ucsf.
edu (W.C.G.)

Table 1. FV infection characteristics of various mouse strains used in this study.

General FV Neutralizing Cell-mediated Splenomegaly

Type

Strain

susceptibility

Viremia

Rfv3

antibody

H-2 *

immunity

Fv2\

induction

Wild type

C57BL/6 (B6)

Resistant

Resistant

r/r

High

b/b

High

r/r

No

BALB/c

Susceptible

Chronic

s/s

Low

d/d

Very low

s/s

Yes

A. BY

Susceptible

Chronic

s/s

Low

b/b

High

s/s

Yes

129/Ola$

Resistant

Resistant

r/r

High

b/b

High

r/r

No

F t hybrids

B6 x BALB/c

Susceptible

Acute

r/s

High

b/d

Low

r/s

Yes

B6 x A.BY

Susceptible

Acute

r/s

High

b/b

High

r/s

Yes

•H-2 is the murine major histocompatibility complex that dictates cell mediated immunity against FV (S, IS). f Fv2 is a dominant FV susceptibility gene that facilitates splenomegaly
induction through aberrant signaling in erythroblasts (4). fFV susceptibility data on 129/Ola were based on results from this study (Fig. 20 (74). The cell-mediated immune response of this
strain was inferred from its H-2 haplotype.

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1343

REPORTS

early control of FV infection in adult immuno-
competent mice.

Rfv3 -mediated recovery from FV infection
correlates strongly with FV-specific neutralizing
antibody responses at 28 dpi (7). Thus, mA3* and
mA3 Fi congcnic strains were monitored for up
to 1 month after FV infection. However, Fj mice
lacking the B6 mA3 allele suffered a markedly
higher rate of FV- induced death (Fig. IB),
indicating that, like Rjv3* susceptibility, mA3
inactivation compromised recovery from FV dis-
ease. Compared with mA3* Fi mice, the three
surviving mA3 Fi mice exhibited mean viremia
higher by a factor of 14 (fig. S2A) and low FV-
specific neutralizing antibody titers at 28 dpi (fig.
S2B), but the small number of surviving animals
precluded obtaining statistically significant data
Therefore, separate cohorts of mice were studied
for FV-specific antibodies at 14 dpi, before the
steep decline in survival of mA3 Fj mice. The
mA3 Fj mice exhibited significantly less FV
binding antibody than mA3* Ft mice, and the
low levels of FV antibodies in mA3 Fi mice
proved comparable to levels detected in the
parental RJv3 3/l BALB/c mice (Fig. 1C). These
findings indicated that mA3 influenced FV-
specific antibody responses.

To better assess FV-specific neutralizing
antibody responses in mice expressing or lacking
mA3, these studies were repeated in high-
recovery B6 x A.BY Fj mice, which generally
survive more than 1 month after FV infection
because of protective cell-mediated immune
responses associated with the H-2^ haplotypc
(Table 1) (5, IS). Plasma samples obtained at 28
dpi revealed significantly lower FV-specific
neutralizing antibody titers in mA3 F, mice

compared with mA3* Fi mice (Fig. ID). These
findings confirmed that mA3 influenced FV-
specific neutralizing antibody responses and
demonstrated that this effect operated indepen
dently of H-2, a known property of Rfv3 (7).

Purebred B6 mice arc highly resistant to FV
infection (Table 1), but their resistance can be
overcome by inoculating aged mice with high
doses of FV (16) or by using immunodcficicnt
mice, including those that fail to produce specific
antibodies (9). Genetic inactivation of mA3 in B6
mice might therefore recapitulate the Rfv3? sus-
ceptible phenotype without a requirement for
outcrossing to susceptible strains. To test this
possibility, >16 week-old B6 mA3* A and m/1 .T'*'
mice were infected with FV. Plasma viremia was
6.2 times as high in mA3~ h mice as in mA3 '
mice at 8 dpi (Fig. 2A). Furthermore, mA3~
mice exhibited significantly lower neutralizing
antibody titers than wild-type mice by 28 dpi
(Fig. 2B). Thus, mA3 inactivation was sufficient
both to enhance viremia and to diminish neu-
tralizing antibody production even in the highly
resistant B6 genetic background. These results
were confirmed in a second highly resistant strain
of mice, 129/Ola (Table 1 and Fig. 2 C) (14).
Together, these findings demonstrate that genet
ic inactivation of mA3 recapitulates the Rfv?
phenotype and indicate that Rfv3 is encoded
by mA3.

Both resistant and susceptible mouse strains
contain the mA3 gene and express mA3 mRNA.
Cloning of splenocyte mA3 mRNA from the
Rjv3 129/Ola strain revealed the predominant
expression of a full-length m.43 transcript, where-
as most mRNA transcripts from Rfv3 r/r B6 mice
lacked exon 5 sequences (figs. S3 and S4) (17).

mA3 transcripts from both RJv3^ strains BALB/c
and A.BY were distinguished by the absence of
exon 2 sequences (Fig. 3A and fig. S4). Quan-
titative reverse transcription polymerase chain
reaction (RT PCJR) revealed similar levels of
total mA3 mRNA in both the Rfvf /r and RfvS^
strains but a level lower by a factor of 17 of
Exon2-containing transcripts in both RJv3 s/t
mouse strains (Fig. 3B). Thus, the presence of
an alternatively spliced mA3 mRNA lacking
exon 2 coirelatcd with the R/v3’-susccptible
phenotype.

Translational read-through of the mA3 mRNA
lacking exon 2 predicts two possible outcomes:
no mA3 protein expression or a truncated mA3
protein (Fig. 3A and fig. S5). Without suitable
mA3 antibodies, these outcomes cannot be
distinguished. We investigated the antiviral activ-
ity of this latter Aexon 2 mA3 protein by cloning
it as a fusion with an N- terminal triple FLAG
epitope tag. NIH3T3 cells were cotransfected
with the expression cons tract and an FV molec-
ular clone (pLRB302) (18) to test the infectivity
of the resulting FV virions. For comparison,
cotransfections were performed with a FLAG-
tagged expression construct containing the mA3
cDNA from B6 mice lacking exon 5. When
standaidized for relative mA3 expression levels,
the BALB/c Aexon 2 mA3 protein was at least 3
to 5 times less potent at inhibiting FV infectivity
than was the B6 Aexon 5 mA3 protein (Fig. 3C).
In control experiments, the full-length mA3
protein from 129/Ola also potently inhibited FV
(fig. S6). These data indicated that, even if a
truncated mA3 protein were expressed in RJv3 t/s
mice, its antiviral activity would be significantly
impaired.

Fig. 1. The RJv3 genetic restriction is mediated by mA3 in vivo. (A to C) mA3 confers the Rfv3 phenotype
in low-recovery H-2 bd mice. Congenic mA3* and mA3~ (B6 x BALB/c) Fi mice were infected with 140
spleen focus-forming units (SFFU) of FV. (A) mA3 is required for viremic control at an early time point (7
dpi); (B) mA3 inactivation compromises fi/V3-associated survival from FV-induced disease; (O mA3
influences FV-specific immunoglobulin G (IgG) production 14 dpi. FV binding IgG was measured by flow
cytometry using FV antigen-expressing FBL-3 cells. (D) mA3 confers the Rfv3 phenotype in high-recovery
H-2 U1> mice. Congenic (B6 x A.BY)F T mice were infected with 1400 SFFU of FV. Mean 28 dpi neutralizing
antibody (NAb) titers (75% inhibitory concentration, IC 75 ) are significantly lower in mA3~ Fj mice. Open
circles indicate individual mice data, gray bars indicate means, and dashed lines refer to the assay
detection limit. Statistical analyses were performed with two-tailed Student's t test.

1344

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

REPORTS

The involvement of m/13 in the control of FV fide innate immune factor in vivo. In addition,
viremia before the onset of adaptive anti-FV im mA3 influences the development of virus-specific
mime responses confirms its stature as a bona neutralizing antibody responses, perhaps by (I)

A

p* 0 013

B

CL 10 5 -
•o

r

i

1 p - 0 035

1 1# *-
S

| 10 1 -

m

o

o

ooo

if

1"

fr

r

o

1

O

1 10 2 .

I a "

oo

ood&5ooo

B6 mA3 36 mAt

B6 mAT~

B6 mA3

Fig. 2. mA3 influences FV-specific
neutralizing antibody responses in
highly resistant (fv2 r/r ) mice. (A
and B) mA3*'+ and mA3 w ~ B6
mice (>16 weeks old) were infected
with 5000 SFFU of FV. (A) mA3
influences early viremic control at

8 dpi; (B) mA3 is required for FV-
specific neutralizing antibody pro-
duction 28 dpi. (0 mA3 influences

C

1 ,0_

1 '

p*0S4

p-OOIV

<b

S-a. # -

I s •-
is

jr

&

7T

; !

"o’

i -

3000

oomrem ococ

can

neutralizing antibody responses in
129/Ola mice. mA3*'+ and mA3’ v ~

i

12401a
mAr •

129>'Cla B« mAr*

MAT* x 129 Ola

B6 mA3-
K 12401*

129/Ola mice were crossed with

mA3 and /nA3"' " B6 mice to generate Ft offspring. FV-specific neutralizing antibody titers (IC 75 ) at
28 dpi with 5000 SFFU of FV are shown. Additional information is found in SOM text and (14).
Statistical analysis was performed using a two-tailed Student's t test

limiting the early viral antigenic load and evad-
ing a form of “high-zone tolerance" (19-21) or
(ii) inhibiting early FV-induced injury of critical
cell types, such as B cells, T cells, or antigen-
presenting cells, required for the development
of FV-specific humoral immunity. However,
mA3 is expressed in B cells and is evolutionarily
related to activation- induced deaminase, an en-
zyme that controls somatic hypermutation and
class-switch recombination in these cells (22).
Thus, mA3 may conceivably be involved in shap-
ing the antibody repertoire.

The human ApobecS family has been im-
plicated in the control of HIV- 1 infection, but
HIV- 1 encodes Vif, which thwarts the actions of
Apobec3G (A3G) and Apobec3F (A3F) (23-26).
Compromised A3G/A3F antiviral activity may
therefore contribute to the generally poor neutral
izing antibody response observed in HIV- 1 infec-
tion (27). Vif antagonists, if and when they arc
available, may enhance the generation of effec-
tive humoral immune responses against HIV-1.
Finally, studies exploring the apparent intrinsic
resistance of individuals who are extensively ex-
posed to HIV-1, yet remain uninfected, have
genetically mapped this phenotype to chromo-
some 22ql2-13 (12), a location distinguished by
a tandem array of the seven human ApobecS
family members. Genome-wide studies of the
entire human Apobec3 locus, with particular
emphasis on functional differences induced by
alternative splicing, arc clearly merited to fully
explore the potential contribution of this gene
family to HIV resistance, neutralizing antibody
production, and disease progression.

Fig. 3. Molecular basis of Rfv3* susceptibility. (A) Aberrant mA3 exon 2 splicing in
Rfv?* mice. If the wild-type mA3 open reading frame (ORF1) is used, frameshift-
induced translational termination will result in a nonfunctional peptide. However, two
start sites in an alternative reading frame (ORF3) may be used to translate a mutant
mA3 protein with a novel N terminus and a 56 amino acid deletion. (B) Decreased
Exon2*, but not total mA3 transcripts, in Rfv3* mice. Quantitative RT-PCR of total
(left) and Exon2~ (right) mA3 splenocyte transcripts in Rjv3 r/ ' and Rfv3^ mouse strains
was performed. Amplification levels were normalized to beta-actin (left) or total mA3
(right). (O Decreased antiviral activity of the mA3 Aexon2 relative to the mA3 AexonS
spliceoform. An FV molecular clone was cotransfected in NIH3T3 cells with FLAG-
tagged mA3 constructs. The infectivfty of harvested virions was assayed in Mus dunni
cells and normalized against reverse transcriptase activity. Cotransfections with FLAG
vector alone were set at 0% inhibition. Expression of mA3 was assessed by anti-FLAG
immunoblotting. Error bars correspond to SD from triplicate experiments.

2

I

2

Z

7

100 -i

SOO ng | ?SO ng | 100 ng

B6mA3\S

IS

SOO ng | JSOng | 100 ng I

BALB/c mA3 \2 I

-Hsp90

*

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1345

REPORTS

References and Notes

1. K. J. Hasenkrug. B. Chesebro, Proc. Natl. Acad. Sci U.S.A
94. 7811 (1997).

2. VH. Li. W. R. Green. ). Virol 80. 5777 (2006).

3. S. Best, P. Le Tissier. G. Towers, J. P. Stoye, Nature 382.
826 (1996).

4. D. A. Persons ef at.. Nat. Genet 23. 159 (1999).

5. W. ). Britt B. Chesebro, ). Exp. Med. 157. 1736 (1983).

6. B. Chesebro. K. Wehrly, in Advances in Comparative
leukemia Research. P. BentveUen. Ed. (North Holland
Biomedical Press. Elsevier. 1978), pp. 69-73.

7. B. Chesebro. K. Wehrly. Proc Natl Acad. Set US. A 76,
425 (1979).

8. D. Doig. B. Chesebro. ]. Exp. Med. 150. 10 (1979).

9. K. J. Hasenknrg, ). Vkol. 73. 6468 (1999).

10. 1C J. Hasenknrg et at.. J. Virol 69. 2617 (1995).

11. H. J. Super et at.. ). Virol 73. 7848 (1999).

12. Y. Kanari et at.. AIDS 19. 1015 (2005).

13. Y. L. Chiu. W. C. Greene, Anna. Rev. Immunol. 26. 317
(2008).

14. Materials and methods are available as supporting
material on Science Online.

G enome sequence changes that altered the
molecular machinery of development
likely facilitated the evolution of unique-
ly human morphological traits (1, 2). Although
these genetic modifications remain largely uniden-
tified, it has long been thought that they included
changes in gene expression due to positive selec-
tion for nucleotide substitutions that altered the
activity of cis-regulatory elements (J). Several cases
of putatively adaptive sequence change, includ-
ing polymorphisms among human populations

'Genomics Division, Lawrence Berkeley National Labora-
tory, Berkeley, CA 94720, USA. J MRC Human Genetics
Unit. Western General Hospital, Edinburgh EH4 2XU, UK.
'United States Department of Energy Joint Genome
Institute, Walnut Creek, CA 94598, USA.

•Present address: Computational and Mathematical Biol-
ogy, Genome Institute of Singapore 138672, Singapore.
fPresent address: Division of Biology, California Institute
of Technology, Pasadena, CA 91125, USA.

JTo whom correspondence should be addressed. E-mail:
emmbin@lbl.gcw (EJW.RJ; james.noonan@yale.edu (J.P.N.)
§Present address: Department of Genetics, Yale University
School of Medicine, New Haven, CT 06520, USA.

15. B. Chesebro. K. Wehrly. ). Exp. Med. 143. 73 (1976).

16. H. C. Van der Gaag. A. A. Axelrad. Virology 177. 837
(1990).

17. S. J. Rulli Jr. ef at..). Virol. 82. 6566 (2008).

18. J. t. Portis, F. J. McAtee. S. C. Kayman . ). Acquit. Immune
Defic. Syndr. 5. 1272 (1992).

19. J. G. Monroe, A Lowy. R. D. Granstein. M. L Greene.
Immunol. Rev. 80. 103 (1984).

20. A Gonraler Femandet C. Milstein. Immunology 93. 149
(1998).

21. Y. T. Kim. G W. Siskind. Chn. Exp. Immunol. 17. 329
(1974).

22. M. Muramatsu et at.. Celt 102. 553 (2000).

23. K. Stopak. C. de Noronha, W. Yonemoto, W. C. Greene.
Mol Celt 12. 591 (2003).

24. M. Marin. K. M. Rose, S. L Kozak. D. KabaL Not Med. 9.
1398 (2003).

25. A M. Sheehy. N. C. Gaddis. M. H. Malirr, Not. Med 9.
1404 (2003).

26. Y. H. Zheng et at.. J. Virol 78. 6073 (2004).

27. D. R. Burton et at.. Nat Immunol. 5. 233
(2004).

and apparently fixed differences between humans
and other primates, have been shown to affect in
vitro promoter or enhancer function in cell line
reporter assays (4-7). However, the impact of
human-specific nucleotide substitutions on the
in vivo activity of developmental regulatory ele-
ments remains obscure.

In vivo analyses of evolutionarily conserved
noncoding sequences have revealed them to be en-
riched in cis-regulatory transcriptional enhancers
that confer specific expression patterns during de-
velopment (8-11). Recent efforts have identified
conserved noncoding sequences that evolved rap-
idly on the human lineage, but it is not known
whether these sequences include regulatory ele-
ments with altered activities in humans (12-15).
Here, we focus on functionally characterizing the
most rapidly evolving human noncoding element
yet identified, which we termed human -accelerated
conserved noncoding sequence 1 (HACNS1) (12).
Although this 546- base pair (bp) element is highly
constrained in all sequenced terrestrial vertebrate
genomes, it has accumulated 16 human-specific

28. We thank the Transgenic Core Laboratory, the Animal
Facility, and 5. Fspineda at the J. David Gladstone
Institutes for technical assistance: L. Evans. J. Portis.

R. Gallo. R. Locksley, and members of the Greene and
Hasenkrug laboratoiy for helpful discussions; and
R. Givens. 5. Cammack. and G. Howard for manuscript
preparation. This work was supported by the NIAID Division
of Intramural Research at MIH to K.J.H. and B.C.. an NIH
R01 Al 065329 to W.C.G.. and an NIH facility grant to the
J. David Gladstone Institutes. Sequences are deposited in
Gen Bank, with accession numbers EU707568 to
EU70757L

Supporting Online Material

www.sde ncema g.or gfcgi /conte nt/ful 1/32 1/58 94/13 43/DC1

Materials and Methods

SOM Text

Figs. SI to S6

Table SI

References

29 May 2008: accepted 24 July 2008
10.112 6/srience.ll61121

sequence changes in the -6 million years since
the human -chimpanzee split (Fig. I A). We evalu-
ated tiie significance of this evolutionary acceler-
ation by means of a test statistic that represents
the log-likelihood, or information theoretic “sur-
prisal,” of observing the human sequence given
the orthologous sequences from multiple ter-
restrial vertebrates. Assuming HACNS1 is under
functional constraint in humans, its rapid diver
gence is highly unexpected given its strong con-
servation in these other species [surprisal test
P value = 9.2 x 10 12 (16)]. This divergence
also significantly exceeds the ~4 substitutions
expected if HACNS1 were evolving at the neu-
tral substitution rate in humans [surprisal test
P value = 1.3 x 10 6 (16)]. One explanation for
this marked acceleration is that HACNS1 has
undergone several instances of positive selection
during human evolution that may have altered its
function.

To test this hypothesis, we evaluated the
ability of HACNS1 and its orthologs from
chimpanzee and rhesus macaque to function as
transcriptional enhancers during development,
using a transgenic mouse enhancer assay in
which the activity of each sequence is assessed
through a p-galactosidase (lacZ) reporter gene
coupled to a minimal Hsp68 promoter (17). We
initially examined the potential enhancer activity
of HACNSJ at embryonic day 11.5 (El 1.5). We
tested a 1.2-kb DNA fragment encompassing
HACNS1 that also contained nonconscrvcd se-
quences flanking the element, in order to include
possible functional sequences near HACNSJ not
detected by conservation (table SI ). At El 1 .5, the
human element drove strong and reproducible
reporter gene expression in the anterior limb bud,
pharyngeal arches, and developing ear and eye,
which suggests that HACNSJ acts as a robust
enhancer during development (Fig. 1, B and C,
and fig. SI). In sinking contrast to the highly
reproducible staining driven by the human
enhancer, which extended into the most distal
region of the anterior limb bud in five of six lacZ-
positive embryos (Fig. 1C and HACNSJ embryos

Human-Specific Gain of Function
in a Developmental Enhancer

Shyam Prabhakar, 1 * Axel Visel, 1 Jennifer A. Akiyama, 1 Matak Shoukry, 1 Keith D. Lewis,

Amy Holt, 1 Ingrid Plajzer-Frick, 1 Harris Morrison, 2 David R. FitzPatrick, 2 Veena Afzal, 1
Len A. Pennacchio, 1 * 3 Edward M. Rubin, X3 $ James P. Noonan 1 ^

Changes in gene regulation are thought to have contributed to the evolution of human
development However, in vivo evidence for uniquely human developmental regulatory function
has remained elusive. In transgenic mice, a conserved noncoding sequence (HACNS1) that evolved
extremely rapidly in humans acted as an enhancer of gene expression that has gained a strong
limb expression domain relative to the orthologous elements from chimpanzee and rhesus
macaque. This gain of function was consistent across two developmental stages in the mouse and
included the presumptive anterior wrist and proximal thumb. In vivo analyses with synthetic
enhancers, in which human-specific substitutions were introduced into the chimpanzee enhancer
sequence or reverted in the human enhancer to the ancestral state, indicated that 13 substitutions
clustered in an 81-base pair module otherwise highly constrained among terrestrial vertebrates
were sufficient to confer the human-specific limb expression domain.

1346

5 SEPTEMBER 2008 VOL 321 SCIENCE wvwv.sciencemag.org

REPORTS

1 to 5 in fig. SI), the chimpanzee and rhesus or-
thologs tailed to drive reproducible reporter gene
expression in the distal limb bud, although they
did drive moderately reproducible expression at
the base of the limb (Fig. I , B and C; chimpanzee
enhancer embryos 1, 2, 6, 7, and 8 and rhesus
enhancer embryos 1 to 4, 6 to 8, and 10 in fig. SI).

Two of the embryos that were transgenic for
the chimpanzee ortholog and showed this pattern
also exhibited diffuse, low-level staining that
extended into the anterior limb, which suggests
that the chimpanzee enhancer may possess a
weak capacity to drive expression in this struc-
ture (embryos 6 and 7 in fig. SI). However, this
infrequent pattern was in stark contrast to the
strong and highly reproducible pattern of the
human enhancer. Furthermore, pharyngeal arch,
eye, and ear expression was less reproducible and,
where present, generally weaker in multiple pos-
itive embryos tor both nonhuman orthologs: these
findings suggest additional sites of reduced over-
all enhancer activity relative to the human ortholog.

To assess the HACNS1 limb expression pat-
tern at higher resolution, we sectioned HACNSl
transgenic embryos and found that staining in the
forelimb was restricted to the mesenchyme, form-
ing a continuous expression domain that extended
deep into the limb bud along the anteroposterior

axis at the handplate and shoulder while remaining
more anterior in between (fig. S3). These results
provide evidence that the human -specific se-
quence changes in HACNSJ have resulted in a
gain of function in this otherwise highly conserved
enhancer, increasing its overall robustness and
producing a strong human-specific expression do-
main in the anterior limb bud mesenchyme at
El 1.5. Because the chimpanzee and rhesus or-
thologs yield similar patterns to each other and
show consistent differences relative to HACNSJ,
a parsimonious conclusion would be that the
chimpanzee and rhesus patterns reflect the ances-
tral primate state from which the human-specific
pattern has evolved.

To explore the activity of HACNSJ at a more
advanced stage of limb development, we com-
pared the expression patterns of the human,
chimpanzee, and rhesus enhancers in El 3. 5 trans-
genic mouse embryos. At this stage, the human
element continued to drive reproducible reporter
gene expression in the anterior developing fore-
limb, particularly in the shoulder and the anterior
junction of the forearm and handplate, in 1 1 of 1 2
positive embryos (Fig. 2, A to C; embryos 1 to 1 0
and 12 in fig. S2). In four of these embryos, the
reporter gene activity extended into the future
anterior-most digit of the forelimb (Fig. 2B). Simi-

lar expression, although with weaker staining, was
also ohserved in the corresponding structures in
the hind limb. Imaging of lacZ staining in a
representative JiACNSI transgenic embryo by
means of optical projection tomography [OPT
( J8 )] revealed that the anterior expression evident
in the whole mount extended deep inside the limb
at the forcann-handplalc junction (fig. S4). The
orthologous chimpanzee and rhesus elements
foiled to drive reproducible expression in the distal
limbs at this time point, although a subset of
positive embryos in each case (4 of 10 for
chimpanzee: 3 of 12 for rhesus) showed reporter
gene expression in the shoulder region of the limb
bud, thus recapitulating the proximal tip of the
expression domain of the human enhancer (Fig.
2A: chimpanzee enhancer embryos 2 to 5 and
rhesus enhancer embryos 1, 5, and 6 in fig. S2).
OPT imaging confirmed the absence of repro-
ducible lacZ staining inside the distal limb in
representative embryos transgenic for the rhesus
and chimpanzee enhancers (fig. S4). These results
indicate that the human-specific enhancer activity
persists across multiple developmental stages.
Moreover, they suggest that the robust anterior
limb expression pattern of HACNSJ evolved from
a weaker ancestral pattern that is largely confined
to the base of the limb bud, as evident in the

. W « Chimpanzee

JP ortholog

Fig. 1. Human-specific gain of function in HACNSl. (A) Top: Location
of HACNSl in NCBI build 36.1 of the human genome assembly.
Bottom: Sequence alignment of HACNSl with orthologs from other
vertebrate genomes; positions identical to human are shown in black.
A quantitative plot of sequence conservation is shown in blue above
the alignment (26-28). The location of each human-specific
substitution is indicated by a vertical red line, and the depth of
nonhuman evolutionary conservation at human-substituted positions
is shown by a vertical yellow line that indicates whether each sequence
is identical to chimpanzee and rhesus at that position. The cluster of 13
human-specific substitutions in 81 bp is also indicated. (B) Expression
patterns obtained from the HACNSl enhancer and orthologous
sequences from chimpanzee and rhesus driving expression of a lacZ
reporter gene in E11.5 mouse embryos. Arrows indicate positions in
the anterior limb bud where reproducible reporter gene expression is
present or absent. A representative HACNSl embryo is shown at top to

Rhesus

ortholog

HACNSl ortholog

illustrate the relevant anatomical structures. Three embryos resulting from independent transgene integration events are shown for each orthologous
sequence tested. (C) Number of embryos transgenic for each sequence displaying the limb expression patterns described in the text

www.sciencema 9 .org SCIENCE VOL 321 5 SEPTEMBER 2008

1347

REPORTS

activities of the chimpanzee and rhesus enhancers
at both time points.

We next sought to identify human-specific
sequence changes responsible for the functional
change in the human enhancer. Although the 16
human-specific substitutions within the 546-bp
conserved region corresponding to HACNS1 are
the most striking feature of the 1 ,2-kb ortholo-
gous segments we initially tested for enhancer
function, these segments also included -650 bp
of nonconscrved DNA containing additional
human-chimpanzee sequence differences. To iso-
late the effect of the substitutions within HACNSI
on enhancer function, we synthesized a chimeric
1.2-kb enhancer in which we transferred all 16
substitutions into the chimpanzee sequence back-
ground (16). This “humanized” chimpanzee
enhancer produced an El 1.5 expression pattern
nearly identical to that of the native human en-
hancer; this finding suggests that the human
specific sequence changes within HACNSI are
responsible for the gain of function we observed
(8 of 8 embryos; Fig. 3D and fig. SI). Strikingly,
these human-specific substitutions were signifi-
cantly clustered: 13 of 16 substitutions occurred
within an 81-bp region of the 546-bp conserved
clement [permutation test P value = 1.7 * 10 7
(/6)], which suggests that this region may be
particularly relevant to the human specific func-
tion of HACNSI (Figs. 1A and 3A).

To test this hypothesis, we synthesized a
chimeric 1.2-kb enhancer in which the 13 clus-
tered human substitutions were introduced into
the chimpanzee sequence background (table S2).

At El 1 .5, this clement produced an anterior limb
bud pattern highly similar to the HACNSI pattern
(6 of 6 positive embryos; Fig. 3 and fig. SI). We
also performed the reciprocal experiment, syn-
thesizing a complementary chimeric enhancer
where we replaced the 1 3 human-specific nucleo-
tides in the human enhancer sequence with their
putatively ancestral orthologs from chimpanzee.
This “reverted” enhancer yielded a pattern very
similar to the chimpanzee and rhesus enhancer
patterns shown in Fig. IB, with expression in the
anterior limb bud greatly reduced or absent (Fig.
3 and fig. SI). These results confirm the robust
ness of the functional differences we observed
between HACNSI and its chimpanzee and rhesus
orthologs, and they indicate that the HACNSI
anterior limb bud pattern is largely attributable to
one or more of the 13 clustered human-specific
substitutions we identified. To further dissect the
functional contribution of these substitutions, we
introduced independent groups of six substitu-
tions and three substitutions into the chimpanzee
enhancer sequence (fig. S5). These enhancers
drove variable expression in the anterior limb
bud, which suggests that at least two human-
specific substitutions are required for the gain of
function in HACNSI.

The precise molecular mechanism by which
the substitutions in HACNSI confer the human
specific expression pattern remains to be de-
termined. Computational analysis of predicted
transcription factor binding sites in HACNSI and
its nonhuman orthologs suggested that multiple
sites have been gained and lost in this enhancer

during human evolution (fig. S 6). Predicted human
specific binding sites for the developmental
transcription factors PAX9 and ZNF423 may
contribute to HACNSI enhancer activity, given
that die known expression pattern of PAX9 in the
mouse limb overlaps ihc human-specific limb
domain of HACNSI at El 1.5 and El 3.5, and
ZNF423 is expressed in the mouse handplatc
mesenchyme from El 0.5 ihrough E12.5 (19-21).

Multiple lines of evidence suggest that the
functional changes in HACNSI are due to adaptive
evolution. The rate of human-specific accelerated
evolution in HACNSI is more than 4 times the
local neutral rate. Moreover, this rapid evolution
cannot be explained purely on the basis of biased
gene conversion (BGQ, a neutral mechanism
postulated to cause hotspots of accelerated evolu-
tion in the genome by increasing the local fixation
rate of AT — » GC substitutions (22, 23). Under the
neutral BGC hypothesis, one would expect an
increase in the overall substitution rate across the
entire region of increased AT — » GC substitution
(23). An excess of AT — * GC substitutions is in-
deed present in HACNSI [binomial test lvalue =
1.1 x 10 4 (/<$)], and the element lies in a ~5-kb
genomic region enriched in such substitutions
(Fig. 4). However, the human-specific substitution
rate is elevated only in the narrow 8 1 -bp region in
HACNSI described above and is close to the local
average outside of this window (Fig. 4). These
data, coupled with the human-specific functional
changes in HACNSI, argue against a purely
neutral explanation for the rapid evolution of this
clement in humans.

Fig. 2. Gain of function in
HACNSI persists at E13.5. (A)
Expression patterns obtained
from HACNSI and its chim-
panzee ortholog in E13.5
mouse embryos. Three em-
bryos resulting from indepen-
dent transgene integration
events are shown for each
construct. Close-up views of
forelimb and hindlimb ex-
pression in a representative
embryo for each construct are
shown at left and arrows in-
dicate positions where limb ex-
pression is present or absent
(B) Dorsal view of reporter
gene expression in the distal
anterior forelimb of a HACNSI
E13.5 transgenic embiyo. Ar-
row indoles the most anterior
digit (C) Number of embryos
transgenic for each construct
that display the limb expres-
sion patterns described in the
text

1348

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

REPORTS

Our results evoke the hypothesis that human
specific adaptive evolution in HACNS1 has con-
tributed to uniquely human aspects of digit and
limb patterning. The dexterity of the human hand
is due to morphological differences compared
with other primates that include rotation of the
thumb toward the palm and an increase in the
length of the thumb relative to the other digits (/).
Human -specific changes in hindlimb motpholo-

gy, such as the characteristic inflexibility and
shortened digits of the human foot, facilitated
habitual bipedalism. The gain of function in
HACNSJ may have influenced the evolution of
these or other human limb features by altering
the expression of nearby genes during limb dc
velopment. HACNS1 is located within an intron
of CENTG2, which encodes a guanosine tri
phosphatase activating protein involved in the

13 substitutions s

I Full anterior Imb sUIntng
[~] Reduced staining in limb
[~~| No reproducible limb staining

Fig. 3. Identification of human-specific sub-
stitutions contributing to the gain of function
in HACNS1. (A) Alignment of HACNS1 with
orthologous sequences from other vertebrate
genomes, focused on an 81-bp region in the
element that contains 13 human-specific sub-
stitutions. The position of each substitution is
indicated by a red box above the alignment;
each human-specific nucleotide is highlighted
in red. Positions in the nonhuman genomes that are identical to the human sequence are displayed as
dots. (B) Expression pattern of a synthetic enhancer in which the 13 human-specific substitutions (red box)
are introduced into the orthologous 1.2-kb chimpanzee sequence background (black bar). (C) Expression
pattern of a synthetic enhancer obtained by reversion of these substitutions (black box) in the human
sequence (red bar) to the nucleotide states in chimpanzee and rhesus. (D) Number of embryos transgenic
for each synthetic enhancer that show full, partial, or no expression in the limb at E11.5.

Synthetic enhancer

Fig. 4. Human-specific substitution rate
and proportion of AT -» GC substitutions in
HACNS1 and flanking genomic region. The
rate of human-specific substitutions (solid
red line) and the fraction of human sub-
stitutions that are AT -» GC events (solid
blue line) were estimated in sliding win-
dows across a 9-kb interval around HACNS1.
The average values of each metric for 1 Mb
of genomic sequence centered on HACNS1
are shown for reference (dashed lines). The
546-bp interval corresponding to HACNS1 is
highlighted in yellow.

regulation of endosome function, The next-
closest gene is GBX2, which is located -300 kb
downstream of HACNSJ and encodes an essen-
tial developmental transcription factor (24, 2S).
The role of CENTG2 in limb development has
not been evaluated. Mouse Gbx2 is expressed
in the developing limb, but Gbx2 null mice
have not been described as showing abnormal
limbs (25). The potential impact of human-
specific changes in the expression of these genes
on limb development thus remains to be ex-
plored. We also note that the HACNS1 expres-
sion pattern in transgenic mice may not entirely
recapitulate the precise HACNSJ expression pat
tern in the human embryo. We therefore cannot
rule out the possibility that the accelerated
evolution of HACNSJ reflects selection for
changes in structures other than, or in addition
to, the limb. Elucidating the role of HACNS1 in
human morphological evolution requires further
lines of evidence, including the analysis of GBX2
and CENTG2 expression during human devel-
opment and the generation of JJACNS1 tar-
geted replacement mice. Independent of these
considerations, our study suggests that adapt-
ive nucleotide substitution altered the function
of a developmental enhancer in humans, and il-
lustrates a strategy that could be used across
the genome to understand at a molecular level
how human development evolved through cis-
regulatory change.

References and Notes

1 E. Trinkaus. in The Cambridge Encyclopedia of Human
Evolution. S. Jones. R. Martin. D. Pibeam, Eds.
(Cambridge Univ. Press. New York. 1993). pp. 346-349.

2. S. 8. Carroll. Nature 422. 849 (2003).

3. M. C. King. A. C. Wilson, Science 188. 107 (1975).

4. C. Toumamille et ol.. Nat. Genet 10. 224 (1995).

5. M. V. Rockman et at. PloS Biol. 3. e387 (2005).

6. 5. A. Tishkoff et at. Not. Genet. 39. 31 (2007).

7. G. A. Wray, Not Rev. Genet 8. 206 (2007).

8. a Baffelti. M. A. Nobrega. E. M. Rubin. Nat Rev. Genet
5. 456 (2004).

9. M. A. Nobrega. I. Ovcharenko. V. Altai. E. M. Rubin,
Science 302. 413 (2003).

10. L A. Pennacchio et at. Nature 444. 499 (2006).

11. A. \Asel et ol.. Nat Genet. 40. 158 (2008).

12. 5. Prabhakar, J. P. Noonan. 5. Pa’Sbo. E. M. Rubin,
Science 314. 786 (2006).

13. K. S. Pollard et at. Nature 443. 167 (2006).

14. C. P. Bird et ol.. Genome Biol. 8. R118 (2007).

15. E. C. Bush. B. T. lahn. BMC EvoL Biol. 8. 17 (2008).

16. See supporting material on Science Online.

17. R. Kothary et at. Development 105. 707 (1989).

18. J. Sharpe et at. Science 296, 541 (2002).

19. A. Neubuser. H. Koseki. R. Balling, Dev. Biol. 170, 701
(1995).

20. H. Peters. A. Neubuser. K. Kratochwil. R. Balling, Genes
Dev. 12. 2735 (1998).

21 S. Warming. T. Sutuki. T. P. Yamaguchi, N. A. Jenkins.

N. a Copeland. Oncogene 23. 2727 (2004).

22. K. 1 Pollard et at. PloS Genet. 2. el68 (2006).

23. N. Galtier. L Duret Trends Genet 23. 273 (2007).

24. Z. Nie et at..}. Biol. Chem. 277. 48965 (2002).

25. K. M. Wassarman et at. Development 124, 2923 (1997).

26. M. Blanchette et at. Genome Res. 14. 708 (2004).

27. A. Siepel et at. Genome Res. 15. 1034 (2005).

28. W. J. Kent et at. Genome Res. 12. 996 (2002).

29. We thank members ol the Pennacchio and Rubin
laboratories for critical comments on the manuscript.
Research was done under Department of Energy Contract

www.sciencema 9 . 0 rg SCIENCE VOL 321 5 SEPTEMBER 2008

1349

REPORTS

DE-AC02-0SCH11231. Univereity of California.

L 0. Lawrence Berkeley National Laboratory, and
supported by National Heart. Lung and Blood Institute
grant H 1066681 and National Human Genome Research
Institute grant HG0O3988 (LAP.); the Agency for
Science. Technology, and Research of Singapore (S.P.);
an American Heart Association postdoctoral fellowship

(A.V.); and NIH National Research Service Award
fellowship 1-F32 GM074367 and the Department of
Genetics. Vale University School of Medicine Q.P.N.).

Supporting Online Material

www. sd encem ag.org/cgi/con tent/TuH/321/589 4/13 46/DC1
Materials and Methods

Figs. SI to S6
Table SI
References

2 May 2008: accepted 7 July 2008
10.112 6/science.US9974

Wnt3a-Mediated Formation of
Phosphatidylinositol 4,5-Bisphosphate
Regulates LRP6 Phosphorylation

Weijun Pan, 1 * Sun-Cheol Choi, 2 * He Wang, 3 * Yuanbo Qin, 3 Laura Volpicelli-Daley, 4
Laura Swan, 4 Louise Lucast, 4 Cynthia Khoo, 5 Xiaowu Zhang, 6 Lin Li, 3 Charles S. Abrams, 5
Sergei Y. Sokol, 2 Dianqing Wu 3 f

The canonical Wnt-p-catenin signaling pathway is initiated by inducing phosphorylation of one of the Wnt
receptors, low-density lipoprotein receptor-related protein 6 (LRP6), at threonine residue 1479 Ohr 1479 ) and
serine residue 1490 (Ser 1490 ). By screening a human kinase small interfering RNA library, we identified
phosphatidylinositol 4-kinase type II a and phosphatkJylinositol-4-phosphate 5-kinase type I (PIP5KI) as
required for Wnt3a-induced LRP6 phosphorylation at Ser 1490 in mammalian cells and confirmed that these
kinases are important for Wnt signaling in Xenopus embryos. Wnt3a stimulates the formation of
phosphatidylinositol 4,5-bisphosphates [Ptdlns (4,5)P 2 | through frizzled and dishevelled, the latter of which
directly interacted with and activated PIP5KI. In turn, Ptdlns (4,5)P ? regulated phosphorylation of LRP6 atfhr 1479
and Ser 149C . Therefore, our study reveals a signaling mechanism for Wnt to regulate LRP6 phosphorylation.

M embers of the Wnt family of secretory
glycoproteins have important roles in
various physiological and pathophys-
iological processes, including embryonic de-

velopment, bone development, neuionogenesis,
adipogenesis, myogenesis, organogenesis, lipid
and glucose metabolism, and tumorigenesis (1-5).
Canonical Wnt binds to two receptors, lipoprotein

receptor-related protein 6 (LRP6) and Sizzled (Fz)
proteins, leading to phosphorylation of LRP6 at
Thr 1479 by casein kinase ly and at Ser 1490 by
glycogen synthase kinase 3 (GSK3) (6-10). Wnt
appears to regulate Thr 479 phosphorylation by
inducing the formation of LRP6 aggregates (9),
whereas it regulates Ser 1490 phosphorylation
through GSK in an axin- dependent manner (10).
To determine whether there are other kinases that
take part in the regulation of LRP6 phospho-
rylation, we screened a human kinase small in-

'Qepartment of Pharmacology, Yale University School of
Medicine, New Haven, CT 06510, USA. 2 0epartment of
Developmental and Regenerative Biology, Mount Sinai School
of Medicine, New York, NY 10029, USA. *State Key laboratory
of Molecular Biology and Center of Cell Signaling, institute of
Biochemistry and Cell Biology, Shanghai institutes for Bio-
logical Sciences, Chinese Academy of Sciences* Shanghai
200031, China. '’Department of Cell Biology and Howard
Hughes Medical institute, Yale University School of Medicine,
New Haven, CT 06510, USA. 5 Department of Medicine,
University of Pennsylvania, Philadelphia, PA 19104, USA. ‘Cell
Signaling Technology, Danvers, MA 01923, USA
•These authors contribute equally to this work,
flo whom correspondence should be addressed. E-mail:
dan.wu@yale.edu

X(V-c.*t«nin

IlL

Fig. 1. Effect of depletion of Ptdlns kinases on
Wnt3a signaling. (A and B) Effects of Ptdlns
kinase siRNAs on Wnt3a-induced phosphorylation
of LRP6 at Ser 1490 . HEK293T cells were trans-
fected with siRNAs, as indicated, for 48 hours and
then treated with Wnt3a (50 ng/ml) for 30 min.
Phosphorytated proteins were assayed by Western
blotting. The experiments were repeated at least
three times. Representative images are shown.
The asterisk in (A) indicates nonspecific bands.
(C and D) Control MO oligos (Ctr MO, 10 nM)
or MO (10 nM) targeting (C) Xenopus PI4Kllot
or (D) PIPSKIa or PIPSKip was injected with
XWntS (2 pg) or xp-catenin (10 pg) mRNA into
four-cell stage embryos, n > 40 for all of the Xenopus embryo studies (where n is the number of examined embryos). Open bars, no double axis; light gray bars,
incomplete double axis; black bars, complete double axis. (E) Four-cell stage embryos were injected with XPIP5KIP MO (40 ng), XPI4Kllct MO (40 ng), or XPIPSKla
MO (40 ng) with or without XPIP5Kip (10 pg) or XPI4Klla (5 pg) RNA in the dorsal region and cultured to tailbud stages. XPIPSKla MO, n = 30; XPIP5Kip
MO, n m 45; XPIP5Klp MO+XPIP5Klp, n = 29; XPI4Klla MO, n = 55; and XPI4Klla MO+XPI4Klla, n = 30.

1350

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

REPORTS

terfering RNA (siRNA) library from Applied
Biosystems for effects on Wnt-induced accu-
mulation of cytosolic p-catenin detected by an
enzyme-linked immunosorbent assay (ELISA) and
on phosphorylation of Ser 1490 of LRP6 detected by
protein immunoblotting in human embryonic kidney
(HEK) 293T cells {11). Multiple phosphatidylinositol
(Ptdlns) kinase siRNAs inhibited cytosolic P-catenin
accumulation (fig. SI A and table SI ) and the phos-
phorylation of LRP6 at Ser 1490 (Fig. 1, A and B) in
response to purified Wnt3a protein. Among the
tested Ptdlns kinase siRNAs, siRNAs tor phospha
tidylinositol 4-kinase type II a (PMKIIa) and
phosphatidylinositoM -phosphate 5-kinase type
I p (PIP5KJP) liad the strongest inhibitory effects
(Fig. 1, A and B, and fig. S1A). These siRNA also
inhibited Wnt3a-induced reporter gene activity (fig.
SIB). Additional siRNAs tor both PMKIIa and
PIP5KIP directed against different targeting se-
quences were also tested (Fig. 1 , A and B, and fig.
SI , B and Q. To further verify that the effects of
these siRNAs were specific, we restored Wnl sig-
naling by expressing the kinases knocked down by
the siRNAs (fig. S I , D to G). PI4KIIa and PIP5KIp
siRNAs did not inhibit lithium- and axinl/2 siRNA
induced accumulation of p-catenin, but p-catenin

siRNA did (fig. SI, C and H). Thus, these siRNAs
appeared to affect Wnt signaling by affecting LRP6
phosphorylation rather titan Wnt signaliqg down-
stream components

PIP5KF/ siRNAs also showed a weak inhibi-
tory effect, whereas PEPSKla siRNAs had no
effect in HEK293T cells (Fig. IB). However, treat
ment of the cells with combinations of the PIP5KI
siRNAs showed that the combination targeting all
three PIP5KI isoforms reduced \Vnt3a-indcued
accumulation of p-catenin and phosphorylation of
LRP6 almost to basal levels (fig. SI, C and I),
suggesting that PIP5KIa and ly may also con-
tribute to Wnt signaling in these cells.

We next examined the roles of the Ptdlns
kinases in Wnt signaling with the use of Xenopus
embryos. A morpholino (MO) targeting Xenopus
PI4KIIa inhibited XWnt8-induced, but not P-
catenin-induced, axis duplication in Xenopus em-
bryos (Fig. IC and fig. S2A). Although PIP5KJP
MO showed little effect, a MO targeting its close
homolog PIP5Kla inhibited XWntS -induced, but
not p-catenin-induced, axis duplication (Fig. ID).
Consistent with the phenotypes, PIP5KIa and
PMKIIa MOs reduced phosphorylation of LRP6
(fig. S2B). In addition, the expression of catalyt-

pS. 490^11 1, ft •

LRP5/1

Wnt3a:

■■■■■■

siRNA: O' 4 ti^Cl 4 t^Cp
pS1490- M

LRP6

4KII<i- « - - — — —

Wnt3« ♦ ♦ ♦ ♦

♦ ♦ ♦

PI(4)P ♦ ♦ ♦ ♦

c

siRNA: oVV/t? ^
pSi490- — m**

LRP(> „.r-

Wnl3a ♦ ♦ ♦ ♦

Pt(4)P «. «. «.

PK4.5)P i : ♦ ♦ ♦ ♦

D

* 150

£ 100

f 50

i o

FRB* FRB-
FK-IP FKBP

Rapamycin:

FRP*FKBP:

FRPiFK-IP

LRP5/6 - HU

Wnt3a ♦ ♦ ♦ ♦

Fig. 2. Effect of Ptdlns (4,5)P 2 on Wnt3a signaling. (A) Effect of exogenous Ptdlns (4,5)P 2 on Wnt3a-induced
phosphorylation of LRP6 at Ser 1490 . HEK293T cells were treated with various Ptdlns lipids in a lipid carrier for
10 min and incubated with Wrrt3a (20 ng/mt) for an additional 20 min before being assayed by immuno-
blotting. (B and O Rescuing the effects of PI kinase siRNAs by direct delivery of Ptdlns lipkk. The asterisk in
(B) indicates nonspecific bands. (D and E) Reduction in Ptdlns (4,5)P 2 levels deceases LRP6 Ser 1490
phosphorylation. HEK293T cells transfected with FRB (PM-FRB-CFP), FKBP (mRFP-fKBP12), or FK-IP (mRFP-
FKBP12-5-ptase-dom) were treated with Wnt3a (20 ngAnl) in the presence or absence of rapamycin (100 nM)
for 30 min before they were collected for the lipid assay (D) and immunoblotting analysis (0. *P < 0.01
compared with the same transfection in the absence of rapamycin (Student's t test). Error bars indicate SDs.

ically inactive PIPSKIa and PMKIIa mutants sup-
pressed axial duplication induced by XWnt8, but
not by p-catenin (fig. S2C), further confirming
the importance of these kinases in Wnt signaling.

Inhibition of zygotic Wnt-p-catenin signaling
induces antcriorized phenotypes that include en-
larged cement glands and head structures (12).
PMKIIa MO that was injected into the dorsal re-
gions of Xenopus embryos induced strong antc-
riorized phenotypes in more than 70% of embryos.
This effect could be partially reversed by co-
injection of Xenopus PMKIIa mRNA (Fig. IE
and fig S2D). Although PIP5KIa MO had Utile
effect on the phenotype, PIP5KIP MO induced
anteriorized phenotypes in -50% of the treated
embryos. The PIP5KIp MO effect could be almost
completely reversed by the injection of XPIP5KIP
mRNA (Fig IE and fig. S2D). Together, these re-
sults indicate that Ptdlns kinases regulate endog-
enous Wnt signaling in Xenopus embryos.

Because sequential phosphorylation of Ptdlns
Upids by PHKH and PIP5KI constitutes the major
pathway for Ptdlns (4,5jP 2 production in most cells
(13, 14), we suspected that Ptdlns (4,5jP 2 might
regulate the phosphorylation of LRP6 at Ser' 490 . To
test this hypothesis, we delivered Ptdlns, plus all of
the seven possible isofoims of Ptdlns phosphates at
equal molar concentrations, into HEK 293T cells in
a lipid carrier. Ptdlns (4,5)P 2 showed the strongest
stimulatory' effect on Wnt3a-induced phosphoryl-
ation of Ser 1490 (Fig. 2A). In addition, the delivery’
of Ptdlns (4)P (but not Ptdlns) rescued the effect erf
PI4K siRNA, whereas the delivery of Ptdlns (4,5)P 2
(but not Ptdlns (4)P) rescued the effect of PIP5K
siRNA on Wnt3a-induced phosphorylation of
LRP6 (Fig. 2, B and C) and P-catenin accumula
tion (fig. S3). These results suggest that Ptdlns

(4,5)P 2 may be the primary’ Ptdlns lipid involved
in the regulation of Ser 149 phosphorylation.

We used a rapidly inducible Ptdlns (4,5)P 2 hy-
drolysis system to further investigate the role of
Ptdlns (4,5)P 2 in regulating phosphorylation of
LRP6. In this system, rapamycin induces the
heterodimerization of membrane -targeted FRP
(fragment of mammalian target of rapamycin that
binds FKBP 12) and FKBP 12 (FK56(Tbinding
protein 12) (used with a truncated form of type IV
phosphoinositide 5-phosphatase, leading to acti-
vation of the phosphatase (15, 16). As shown in
Fig. 2D, rapamycin reduced the amount of Ptdlns

(4,5)P 2 in cells expressing both FRP and phosphatase-
tused FKBP12. but not in those expressing FRP
and FKBP 12 alone. Rapamycin also attenuated
phosphorylation of LRP6 only in cells expressing
both FRP and phosphatase fused FKBP12 (Fig. 2E).

To determine whether Wnl3a can stimulate
Ptdlns (4,5)P 2 production, we established a Ptdlns

(4,5)P 2 ELISA and detected significait Wnt3a-
induced formation of Ptdlns (4,5)P 2 (more than two-
fold increases) in HEK293T, Hela, and NIH3T3
cells (fig. S4A)i We confirmed these results by high-
performance liquid chromatography (HPLC) and
thin- layer chromatography (TLC) (Fig. 3A and fig.
S4, B and C). Together with the findings that the
PMKIIa siRNA abolished (and the PIP5KIp siRNA

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1351

REPORTS

Fig. 3. Stimulation of Ptdlns (4,5)P 2 formation by
Wnt3a through Fz and DvL (A) Effect of Wnt3a
treatment on Ptdlns (4,5)P 2 content HEK293T Cells
were stimulated with Wnt3a protein (50 ng/ml)
before lipid extraction. Ptdlns (4,5)P 2 content was
determined by HPLC. *P < 0.01 compared with
time 0 (Student's t test). (B) Requirement of PI4Klla
and PIPSKip for Wnt3a-induced formation of Ptdlns

(4.5) P 2 . Cells were transfected with siRNAs, as in-
dicated, for 48 hours and then treated with Wnt3a
(50 ng/ml) for 30 min. Ptdlns (4,5)P 2 were de-
tected by ELISA. (C) Effect of Fz siRNAs on Wnt3a-
induced formation of Ptdlns (4,5)P 2 . Cells were
transfected with control siRNA or a combination of
Fz2, Fz4, and Fz5 siRNAs for 48 hours and then
treated with Wnt3a (50 ng/ml) for 30 min before
assays. *P < 0.01 compared with control siRNA
transfection in the absence of Wnt3a (Student's
t test). (D) Effect of Fz siRNAs on Wnt3a-induced
phosphorylation of LRP6 at Ser 1490 . Cells were
transfected as in (O for 48 hours and then treated
with Wnt3a (50 ng/ml) for 30 min. (E) Effect of Fz
overexpression on accumulation of Ptdlns (4,5)P 2 .

HEK293T cells were transfected with the LacZ, Fz5,
or LRP6 expression plasmids for 18 hours, and Ptdlns

(4.5) P 2 levels were determined by ELISA. *P < 0.01
compared with the sample expressing LacZ (Stu-
dent's f test). (F) Effect of Fz5 expression on phos-
phorylation of LRP6 at Ser 1490 . Cells were transfected
with Fz5 expression plasmid for 18 hours and then
treated with Wnt3a (20 ng/ml) for 20 min. (G) Effect
of Dvl expression on the Ptdlns (4,5)P 2 levels.

HEK293T cells were transfected with the mouse
Dvll, 2, or 3 expression plasmid for 18 hours before
the Ptdlns (4,5)P 2 ELISA assay. *P < 0.01 compared
with the sample expressing LacZ (Student's t test).

(H and I) Effect of CM siRNAs on formation of Ptdlns

(4,5)P 2 and phosphorylation of LRP6 at Ser 1490 .

HEK293T cells were transfected with control siRNA
or Dvl siRNA mixture targeting Dvll, 2, and 3 for
48 hours and then treated with Wnt3a (50 ng/ml)
for 30 min. *P< 0.01 compared with control siRNA
transfection in the absence of Wnt3a (Student's
/test), (p Interaction of Dvl3 with endogenous PIPSKip.

HEK293T cells (Dvl) stably expressing Dvi3-WA 7 were
used in immunopredpitation by an antibody against
HA (anti-HA). The parent HEK293T cells (HEK) were used as a controL Immuno-
complexes were detected by the anti-Dvl3 and anti-PIP5Kip antibodies. (K) Effect of
purified recombinant Dvl3 protein on kinase activity of purified recombinant PIP5K1P
protein. PIPSKip (50 nM) was incubated with the GST (glutathione S-transferase)or

LacZ F i2 F?4 FrSLRPe

LacZ Dvll Dvl2 Dvl3

i

1

-- -Mil

Cell HEK Dvl Dvl HEK Dvl Dvl
Wnt3a. - - ♦ . . ♦

V 1

:

i

j

O -Wnt3a
■ ♦Wnt3a

a

LacZ F?5 LacZ FzS

pSU00 flHHB

siRNA: Ctr Dvl Clr Dvl

Dvta-_

100 200 300 400
Protein (nM)

Dvl3 proteins for 2 hours at room temperature. One-tenth of die samples was taken
for Western blotting, and the rest was subjected to in vitro kinase assay with Ptdlns
(4)P as a substrate. The product Ptdlns (4,5)P 2 is separated by TIC detected, and
quantified by a phosphoimager. Error bars indicate SDs in all panels.

reduced) Wnt3a-induccd accumulation of Ptdlns
(4 t 5)P 2 (Fig. 3BX we conclude that Wnt3a stimulates
Ptdlns (4 v 5jP 2 production via these Ptdlns kinases.

We next investigated whether Fz is required for
WnGa-induced fonnation of Ptdlns (4,5)P2 and
detected expression of Fz2, 3, 4, 5, and 6 in
HEK293T cells by reverse transcription polymerase
chain reaction (table S2). We made and validated
two sets of siRNAs for each of these Fz genes (table
S2) FzS siRNA showed the strongest inhibition of
Wnt3a induced accumulation of p-catenin, whereas
Fz2 and Fz4 siRNAs also had some inhibitory
eftect (fig. S5B and table S2). The combination of
Fz2, 4, and 5 siRNAs virtually abolished Wnt3a-
induced accumulation of P-catenin (table S2). This
combination also abrogated Wnt3a-induced fonna-

tion of Ptdlns (4,5jP 2 (Fig. 3C) and phosphoryl
ation of LRP6 at Ser 1490 (Fig. 3D, fig. S5C, and
table S2). On the other hand, expression ofFz5, 2,
and 4 stimulated die fonnation of Ptdlns (4,5jP 2
(Fig. 3E) and Wnt3a-inducod phosphorylation of
Ser 1490 (Fig. 3F, and fig. S5, D and E). These results
together indicate that Wnt3a acts through Fz to
stimulate Ptdlns (4,5 )P 2 fonnation in HEK293T
cells and regulates LRP6 phosphorylation.

Because dishevelled (Dvl) is required for the
phosphorylation of LRP6 (9, 10 ), we questioned
whether Dvl might have a role in the formation of
Ptdlns (4,5)P 2 . Expression of Dvl 1-3 increased the
amount of Ptdlas (4.5)P 2 in 1IEK293T cells (Fig.
3G and fig. S6A). When HEK293T cells were
transfected with a mixture of three Dvl siRNAs tar-

geting each of the three Dvl isoforms ( 17 ), both
Wnt3a^ induced fonnation of Ptdlns (4,5 )P 2 and
phosphorylation of Ser 1490 were inhibited (Fig. 3, H
and L and fig. S6B). Dvl is a scaffold protein widi
no known enzymatic domains. We tested whether
Dvl aid PIP5KIP interacted and found that they
coimmunoprecipitated when overexpressed in
HEK293T cells (fig. S6C). We mapped Dvl interac-
tion site to the N -terminal half of PIP5KJP kinase
domain (fig. S6D) and PIP5KI-binding sites to two
fragments of Dvll that contain the DIX and PDZ
domain, respectively (fig. S6E). The interaction of
Dvl3 and PIP5KIP was also examined in a
HEK293T cell line that stably expressed Dvl3 car-
rying seven hemagglutinin (HA) tags at its C ter-
minus at a level lower than that of endogenous Dvl3

1352

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

REPORTS

I

2 3 4 9 6 7 8 B 10 11 12

LRP6-
LRP6- __
LRP6- '

LRP6

FRP*FKBP

FRP+FK-IP

Axtnl-

■la

FRP*FKBP
FRP-»FK-IP

p-T 1479-

P-S1490-.

LRP6- r r t-J r-,„

Wnt3a -
Rap

Fig. 4. Requirement of Ptdlns (4,5)P 2 for formation of LRP6
aggregate and membrane translocation of axin and GSK3. (A)
Requirement of Ptdlns (4,5)P 2 for Wnt3a-induced LRP6 aggregation.
HEK293T cells were transfected and treated with Wnt3a and
rapamydn as indicated. Cell lysates were subjected to sucrose
density-gradient ultracentrifugation, and fractions were analyzed by
Western analysis. (B) Ptdlns (4,5)P ? content in two sucrose density-
gradient uttracentrifugation fraction pools. Fractions 8 to 11 and 1 to
4 from (A) were pooled, and Ptdlns (4,5)P 2 amounts were measured
by ELISA. Open bars correspond to the samples from the top panel in
(A); black bars to the second panel; striped bars to the third panel;
and dotted bars to the last panel The Ptdlns (4,5)P 2 amounts are
presented relative to those in untreated cells. Error bars indicate SDs.
(O Requirement of Ptdlns (4,5)P 2 for phosphorylation of LRP6 at
Thr 1479 . HEK293T cells were transfected with plasmids, as indicated,
for 20 hours and then treated with Wnt3a (20 ng/ml) for 30 min in
the presence or absence of rapamycin (100 nM) before they were collected for immunoblotting analysis. The asterisk indicates nonspecific bands. (D) Requirement of
Ptdlns (4,5)P 2 for Wnt3a-induced membrane recruitment of axinl. HEK293T cells were transfected and treated as indicated. The membrane fractions were prepared and
analyzed by Western analysis. (E) Model for Wnt3a cross-membrane signaling.

LRP6-«i m m m m m

Wnl3a

♦♦♦♦♦♦

(upper right panel in Fig. 3J). Although we did not
observe coimmunoprecipitation in the absence of
\Vnt3a, an interaction of Dvl3-HAwith endogenous
PlPSKip was detected in the presence of Wnt3a
(Fig. 3J), suggesting that Wnt3a may regulate the
interaction. In a pull-down assay with recombinant
proteins prepared in E. coU, Dvll interacted with
PlPSKip and la in vitro (fig. S6F). We then tested
whether Dvl could directly regulate PIP5KI kinase
activity. Using the recombinant Dvl3 and PIP5Kip
prepared from E coli, we found that, in an in vitro
kinase assay, Dvl directly stimulated PIP5K1P in a
dose-dependent manner (Fig. 3K and fig. S6G).
Together with the knowledge that Fz can interact
with Dvl and recruit it to the membranes (9, 18-23\
the above data suggest that Wnt3a may induce
(through Fz) Dvl to bind and activate PIP5K1
We next tested whether Ptdlns (4,5)P2 is re-
quired for Wnt3a-induced formation of LRP6
aggregates (referred to as “signalosomes”), which
precedes phosphorylation of LRP6 at Thr 1479 (9).
We used sucrose density-gradient centrifugation to
detect LRP6 aggregates (Fig. 4A). Fractions that
contained LRP6 aggregates also had a higher
Ptdlns (4,5)P 2 content than did fractions containing
nonaggrcgatcd LRP6 (Fig. 4B). Notably, the aggre
gation was sensitive to the elimination of Ptdlns

(4,5)P 2 through rapamycin-induced Ptdlns (4,5)P 2
hydrolysis (Fig. 4A). We also examined LRP6 ag-
gregation using confocal microscopy in He la cells
expressing LRP6 YFP (9). We observed the aggre
gates in control cells but not in cells transfected with
the PIP5KI siRNAs (fig. S7). Elimination of Ptdlns

(4,5)P 2 also led to decreased phosphorylation of

Thr 1479 (Fig. 4C) (9). Therefore, we conclude that
Ptdlns (4,5)P 2 is required for Wnt-induced LRP6
aggregation and Thr 4,9 phosphorylation.

Because I.RP6 aggregates appear to have a high
affinity for axin (9) and because axin membrane
translocation is required for GSK3 mediated phos-
phorylation of Scr 1490 (10), we examined if Ptdlns
(4 r 5)P 2 is involved in Wnt-induced axin membrane
translocation. Elimination of Ptdlns (4,5)P 2 using
the rapamycin-inducible Ptdlns (4,5 )P 2 hydrolysis
system abrogated Wnt3a-induced axin translocation
(Fig. 4D> Putting all of these results together, we
propose a model (Fig. 4E) to suggest that Wnt3a
regulates the activity of PIP5K1 through Fz and Dvl
and induces the formation of Ptdlns (4,5)P 2 . Ptdlns

(4,5)P 2 is required, but not sufficient, for LRP6
aggregation and phosphorylation at both Thr 1479
and Scr 1490 , as well as for axin translocation. Precise
mechanisms by which Ptdlns (4,5)P 2 regulates these
Wnt signaling events, however, need to be inves-
tigated further.

References and Notes

1 C. Y. Logan. R. Nusse. Anna. Rev Cell Dev. Biol. 20. 781
(2004).

2. T. Reya. H. Clevers. Nature 434. 843 (2005).

3. H. (levers. Cell 127. 469 (2006).

4. R. T. Moon. A. D. Kohn. G. V. De Ferrari. A. Kaykas,

Not. Rev Genet 5. 691 (2004).

5. A. Mani et at.. Science 315. 1278 (2007).

6. G. Davidson et at.. Nature 438. 867 (2005).

7. R. Nusse. Nature 438. 747 (2005).

8. X. Zeng ef a!.. Nature 438, 873 (2005).

9. J. Bilic et at.. Science 316. 1619 (2007).

10. X Zeng et a!.. Development 135. 367 (2007).

11. Materials and methods are available as supporting
material on Science Online.

12. R. Hart and. J. Gerhart. Anna. Rev. Cell Dev BioL 13. 611
(1997).

13. J. J. Hsuan, S. Minogue. M. dos Santos. Mv. Cancer Re t
74. 167 (1998).

14. R. L Doughman. A. J. Firestone. R. A. Anderson.

/ Membr. Biol. 194, 77 (2003).

15. W. D. Heo et at.. Science 314. 1458 (2006). published
online 8 November 2006; 10.1126/science.ll34389.

16. P. Vamai, B. Thyagarajan. T. Rohacs. T. Balia. / Celt BioL
175. 377 (2006).

17. L li. J. Mao. L. Sun. W. Liu. D. Wu. }. Biol. Chem. 277,
5977 (2002).

18. J. D. Axelrod. J. R Miller. J. M. Shulman. R. T. Moon.

N. Perrimon, Genes Dev. 12. 2610 (1998).

19. H. C. Wong et at.. MoL Celt 12. 1251 (2003).

20. W. Chen et at.. Science 301. 1391 (2003).

21. F. Cong. L Sdrweirer. H. Varmus, Development 131,
5103 (2004).

22. W. J. Pan e( at.. Cell Rei 14. 324 (2004).

23. N. Yokoyama. D. Yin. C. C. Malbon. /. MoL SignoL 2. 11
(2007).

24. We thank P. De Camilti for making available resources from
Ns Lab in connection to Ihis work, discussion, and critical
reading of the manuscript; Z. li. M. Orsulak. Y. Zhang.

L Tang W. Liu, Y. XL F. Nie. Y. Wu and X Gan for technical
assistance, and D. Sussman. C. Carpenter, R. Grossdtedl
T. Balia, B. Williams M. F. Roussel X. He. C. Nieto*. R. Nusse.
and J. Nathans for providing experimental materials. This work
was supported by grants from NIH (AR051476. CA132317,

H 1080706. NS 36251 and DA018343). the National Institute
on Drug Abuse to Yale Neuroproteomics Center, the Ministry
of Science and Technology of China (2002CB5 13000), and
the National Science Foundation of China (30521005).
Supporting Online Material
www.scie rveema g. or g/cgi/c onte nt/futl/32 1/58 94/13 50/DC 1
Materials and Methods
Figs. SI to S7
Tables SI to S4
References

20 May 2008; accepted 4 August 2008
10.112 6/srience.ll6074l

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1353

REPORTS

Helical Structures of ESCRT-III
Are Disassembled by VPS4

Suman Lata, 1 Guy Schoehn, 1,2 Ankur Jain, 3 * Ricardo Pires, 1 Jacob Piehler, 3
Heinrich G. Gottlinger,'' Winfried Weissenhorn 1 !

During intracellular membrane trafficking and remodeling, protein complexes known as
the ESCRTs (endosomal sorting complexes required for transport) interact with membranes and
are required for budding processes directed away from the cytosol, including the budding of
intralumenal vesicles to form multivesicular bodies; for the budding of some enveloped
viruses; and for daughter cell scission in cytokinesis. We found that the ESCRT-III proteins
CHMP2A and CHMP3 (charged multivesicular body proteins 2A and 3) could assemble in
vitro into helical tubular structures that expose their membrane interaction sites on the outside
of the tubule, whereas the AAA-type adenosine triphosphatase VPS4 could bind on the inside
of the tubule and disassemble the tubes upon adenosine triphosphate hydrolysis. CHMP2A
and CHMP3 copolymerized in solution, and their membrane targeting was cooperatively enhanced
on planar lipid bilayers. Such helical CHMP structures could thus assemble within the neck of an
inwardly budding vesicle, catalyzing late steps in budding under the control of VPS4.

E SCRT (endosomal sorting complexes re-
quired for transport) complexes 0, I, II,
and III and accessory proteins regulate
cell surface receptor sorting into intralumenal
endosomal vesicles, generating multivesicular
bodies (MVBs) (7-3). ESCRTs are also re-
cruited during budding of some enveloped vi-
ruses (4) and cytokinesis (5, 6), processes that
are topologically similar to vesicle budding
into endosomes.

Yeast expresses six ESCRT-III-like proteins
(7), whereas mammalian cells express 10, known
as charged multivesicular body protein (CHMP)
1 to 6 (8). C-terminally truncated CHMP3 has
a four helical bundle core and two regions that
are important for CHMP polymerization and
membrane targeting (9). CHMPs exist in an

auto-inhibited state in the cytosol (10, 11); re
moval of autoinhibition induces membrane tar
geting (9, 12, 13) and ESCRT-III assembly into
a putative protein lattice (7, 2). Overexpression
of SNF7/CHMP4 in mammalian cells produces
filaments that induce outward buds in the pres-
ence of catalytically inactive VPS4 (14), but
little is known about heteromeric polymerization
by ESCRT-III proteins.

The recruitment of the AAA-type adenosine
triphosphatase (ATPase) VPS4 is essential for
the termination of the budding process and cat-
alyzes disassembly of the complex (2, IS, 16).
The central role of ESCRT-III and VPS4 in all
known ESCRT-catalyzed budding events is fur-
ther underlined by the inhibitoiy effects of domi
nant negative mutants of ESCRT-III (8, 9, 17, 18)

and VPS4 (2, 5, 18) on MVB formation, HIV-1
budding, and cytokinesis.

Yeast Vps2p (CHMP2) and Vps24p (CHMP3)
form a subcomplex (7) consistent with the het-
erodimerization potential suggested by the CHMP3
crystal structure (9). Such subcomplcxcs may
thus represent the building blocks for polymer-
ization. We set out to study the interactions and
polymerization mode of CHMP2 and CHMP3
in vitro. Interactions between full-length auto-
inhibited CHMP3 and CHMP2A could not be
detected, and so both proteins were produced as
C- terminal truncations (fig. S1A; CHMP2AAC
and CHMP3AQ in their proposed activated forms
(9, 10, 73). Whereas CHMP3AC was memodis-
perse in solution (9), CHMP2AAC fused to the
maltose binding protein (MBP) formed monomers
and concentration-dependent aggregates (fig.
SIB). Co incubation of monomeric CHMP2AAC
and CHMP3AC led to the formation of oligomers,
migrating to the bottom traction of a sucrose
gradient (Fig. 1 A). CHMP2AAC and CHMP3AC
assembled into long tubular structures (Fig. 1 B),
with the majority of the particles revealing a di-

*Unit for Virus Host Cell interaction, UMR 5233 JJF (Uni-
versite' Joseph Fourier)-EMBL (European Molecular Biology
Laboratory) CNRS, 6 rue Jutes Horowitz, 38042 Grenoble
Cedex 9. France. 2 institut de Biologie Structurale UMR 5075
CEA (Comissariat a I'Energie AlomiqueKNRS-UJF, 41 rue
Jules Horowitz. 38027 Grenoble Cedex 1, France, ’institute of
Biodiemistry, Johann Wolfgang Goethe University, Max-von-
Laue Strafie 9, 60438 Frankfurt am Main, Germany. “Pro-
gram in Gene Function and Expression, Program in Mdecular
Medicine. University of Massachusetts Medical School,
Worcester, MA 01605, USA

* Present address: Department of Physics, University of

Illinois at Uitoana-Champaign, 1110 West Green Street,

Urban a, II 61801-3080, USA

fTo whom correspondence should be addressed. E-mail:

weissenhorn@emW.fr

75

37

25

4 - CHMP2AAC
4 - MBP

4- CHMP3AC

Fig. 1. CHMP2AAC-CHMP3AC polymer formation. (A) Sucrose density
gradient analysis of CHMP2AAC-CHMP3AC complex formation. (B) Negative
staining EM of tubular structures formed by CHMP2AAC and CHMP3AC.
Negative staining EM (C) of CHMP2AAC and (D) of CHMP2AAC after removal
of MBP. Scale bars indicate 100 nm; inset scale bars, 50 nm.

1354

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

REPORTS

Fig. 2. Cryo EM of CHMP2AC-CHMP3AC (A) tubes and (B) tubes after removal
of MBP from CHMP2AAC. Scale bars, 100 nm. (C) The EM reconstruction model
showing the 45 A width of the helical structure produced by the CHMP lattice
(top view).

Fig. 3. CHMP2AAC-CHMP3 tube disassembly by
VPS4B. (A) Negative staining EM of tubes formed
by CHMP2AAC and CHMP3. Sucrose gradient
analysis of (B) CHMP2AAC, (C) CHMP3, (D)
VPS4B, (E) CHMP2AAC-CHMP3-VPS4B complex
formation, and (F) CHMP2AAC-CHMP3-VPS4B
complexes after incubation with ATP Mg 2 *. (G)
Negative staining EM of CHMP2AAC-CHMP3 tubes
revealing VPS4B on the inside. Radial density
profile (H) of a CHMP2AAC-CHMP3-VPS4B and (I)
of CHMP2AAC-CHMP3 tubes calculated across the
cross section of the tube. (J) Negative staining EM
after adding ATP Mg 2 * to CHMP2AAC-CHMP3-
VPS4B tubes. (K) Disassembly of fluorescein-
labeled CHMP2AAC-CHMP3 tubes measured by
change in emission intensity upon addition of HBS
(magenta), 10 pM VPS4B (blue), 10 pM VPS4B
plus 100 pM AMP-PNP Mg 2 * (green), 5 pM VPS4B
plus 50 pM ATP Mg 2 * (black), and 10 pM VPS4B plus
100 pM AP Mg 2 * (red). (Inset) The fluorescein-
labeled CHMP2AAC and CHMP3 visualized on an
SDS-polyacrylamide gel. Scale bars, 100 nm.

www.sciencema 9 .org SCIENCE VOL 321 5 SEPTEMBER 2008

1355

REPORTS

ameter of -40 nm and a small fraction showing
deviations up to 70 nm in negative staining election
microscopy (EM).

Although neither CHMP3, CHMP3AC, nor
CHMP2A fonned higher-order oligomers on their
own, CHMP2AAC sedimented in a sucrose gra-
dient (fig. SIC) and formed ringlike structures
(Fig. 1C). Removal of MBP produced particles
with inner and outer diameters of -12 nm and
-30 nm, respectively. Increased aggregation upon
MBP cleavage and the exclusive tace-up orien-
tations (Fig. ID) hindered determination of the
ring thickness.

CryoEM of the CHMP2 AACCHMP3 AC
polymer showed tubes with a frizzy surface (Fig.
2A). Removal of MBP from CHMP2AAC ren
dered the surface smooth, revealing striations
perpendicular to the longitudinal axis of the tube

(Fig. 2B). Because image analyses of MBP-cleaved
tubes was hindered by their aggregation, we ap-
plied the iterative helical real space recons true
lion algorithm ( 19) to reconstruct volumes from the
tubes containing MBP attached to CHMP2AAC
(20). The 32 A pitch of the helical assembly was
determined by the Fourier transform of the images
(20). The calculated helical structure contains 16.57
repeating units per turn, with inner and outer di-
ameters of 43 ran and 52 ran, respectively (Fig.
2C). The CHMP2A-CHMP3 dimer model could
fit into the repeating unit of the EM map (20) (fig.
S2, A to D), exposing the membrane targeting
surface to the outside and the VPS4B interaction
site toward the inside of the tube (fig. S2, D and E).

The organization of the CHMP2AAC
CHMP3AC polymer did not indicate how other
CHMPs could participate in the same polymer.

In fact, monomeric CHMP4BAC did not integrate
into CHMP2AAC -CHMP3 AC tubes (fig. S3) de-
spite similarities in polymerization (14). Although
CHMP4 may participate in CHMP2A -CHMP3
tube formation as a CHMP4CHMP6 subcomplex
(7\ different ESCRT-1II complexes may be formed
in vivo that interact (7) and act sequentially.

Because the VPS4B CHMP interaction re
quires an intact C terminus (21, 22), we used frill-
length proteins for tube formation. Whereas
CHMP2AAC and CHMP3 fonned similar tubes
to those fonned with CHMP3AC (Figs. IB and
3A) t combinations of CHMP2A and CHMP3
and CHMP2A and CHMP3AC did not polym-
erize. In sucrose gradient centrifugation, isolated
forms of CHMP2AAC, CHMP3, and VPS4B
floated at similar positions (Fig. 3, B to D), con-
sistent with monomeric or dimeric VPS4B in
the absence of adenosine triphosphate (ATP)
(21). Mixing all three proteins recruited them to
the pellet fraction, corroborating the CHMP3
VPS4B interaction (Fig. 3E). Negative staining
EM of the CHMP2AAC CHMP3 tubes assem
bled in the presence of VPS4B revealed additional
electron-dense material along the longitudinal
axis of the tubes (Fig. 3G). Radial density pro-
files of cross sections obtained by cryo-EM (fig.
S4B) showed a broad central peak, confirming
the VPS4B presence inside the tubes, followed
by a sharp peak corresponding to the CHMP lat-
tice and a smaller peak corresponding to MBP
attached to CHMP2AAC (Fig. 3H and fig. S4D).
In contrast, in cross sections of tubes without
VPS4B (fig. S4A), the central peak of the den-
sity profile did not rise above the background
signal (Fig. 31 and fig. S4Q. Thus, VPS4B used
CHMP2AAC CHMP3 tubes as a scaffold tor as-
sembly, which apparently produced disordered
VPS4B oligomers in the absence of ATP. Addition
of ATP and Mg 2 * to the CHMP2AAC CHMP3
VPS4B tubular structures induced disassem-
bly of the tubes (Fig. 3, F and J). Incubation of
fluorescein-labeled tubes (fig. S5) with buffer,
VPS4B, or VPS4B plus adenylyl-imidodiphosphate
(AMP-PNP) Mg 2 * showed no change in emission
intensity as a function of time (Fig. 3K). In con-
trast, dequcnching, indicating disassembly, was
measured as an increase in emission intensity
when the tubes were incubated with VPS4B plus
ATP Mg 2- * (Fig. 3K). Thus. VPS4B induced
tube disassembly in vitro.

CHMPs arc selectively targeted to cellular
membranes (7, 13, 24), which requires an ex-
tended basic surface in case of CHMP3 (9). With
reflectometric interference spectroscopy (RlfS)
(20), CHMP2AAC, CHMP3AC, and their com
plex show'cd no notable mass deposition (<0. 10
ng mm ~) on silica-supported bilayers composed
of l-stearoyl-2-olcoyl^i-glyccro-3-phosphocholinc
(SOPC) lipids (Fig 4A). However, CHMP3AC
bound to 1 ,2-dioleoyl-A7i-glycero-3 phosphoserine
(DOPS):SOPC bilayers (fig. S 6) with an equi
librium binding amplitude (T*,) of 1.08 ng mm 2
and dissociated with a rate of (k^) - 0.3 s -1 ,
whereas CHMP2AAC (4 pM) showed a of

Fig. 4. CHMP protein membrane interaction. (A) Binding curves measured by RlfS on a SOPC bilayen
MBP (black), CHMP3AC (magenta), CHMP2AAC (green), and CHMP2AAC-CHMP3AC (blue). (B) Binding
curves measured by RlfS on a DOPS:SOPC bilayen MBP (black), CHMP3AC (magenta), CHMP2AAC
(green), and CHMP2AAC-CHMP3AC (blue). Protein injections were followed by injection of 1 M NaCl in
HBS. CH MP2A\C-CH MP3 AC tubes (C) with and (D) without MBP attached to CHMP2AAC. (E) CHMP2AAC-
CHMP3AC tubes assembled in the presence of SOPC LUVs. (F) CHMP2AAC-CHMP3AC tubes assembled in
the presence of 0.5 mg ml -1 (left) and 1.5 mg ml -1 (right). SOPCDOPS LUVs are shorter, (G) reveal single
helical coils, and (H) are often cone-shaped. Scale bars, 100 nm.

1356

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

REPORTS

1.36 ng mm 2 and a of 0.081 s _1 (Fig. 4B).
When the two proteins were mixed, a of
5.95 ng mm 2 was reached. An apparent >
0.0008 s _! was determined by fitting a mono-
exponential function to the dissociation phase,
which is 3 and 2 orders of magnitude lower as
compared with the single proteins (fig. S7). Both
SOPC and DOPSrSOPC bilayers showed no in-
teraction with the control protein MBP (Fig. 4,
A and B) Dissociation of bound CHMP2AAC and
CHMP3AC in Hepcs buffered saline (HBS) con-
taining 1 M NaCl revealed a k& > 1 s -1 , which was
much faster than dissociation in HBS alone. In
contrast, die CHMP2AAC-CHMP3AC polymer did
not dissociate with a higher rate (k^ < 0.00046 s~ l )
in the presence of 1 M NaCl (fig. S7), indi-
cating resistance to change in ionic strength.
Once assembled on membranes, CHMP2AAC-
CHMP3AC did not exchange with soluble or
membrane bound CHMP3AC (fig. S8). Thus,
CHMP2A-3 complexes assembled on membranes
in vitro in the absence of CHMP46 subcomplexes,
even though yeast Snf7-Vps20 (CHMP4-6) com-
plexes may recruit Vps2-Vps24 (CHMP2A 3)
complexes to membranes in vivo (7).

To assess the influence of CHMP2AAC-
CHMP3AC tubes on membrane shapes, we
used large unilamellar vesicles (LUVs) com-
posed of DO PS: SOPC. LUV incubation with
either CHMP2AAC or CHMP3AC had no ef-
fect on their floatation in sucrose gradients (fig.
S9, A and B), whereas preformed CHMP2AAC
CHMP3AC tubes restricted LUV floatation to
the middle of the gradient (fig. S9, C and D).
Negative staining EM confirmed CHMP2AAC
CHMP3AC tube membrane interaction via their
outer surfaces (Fig. 4, C and D). However, no
systematic remodeling of the LUV membranes
was observed. Potential membrane remodeling
by the CHMP copolymer or vice versa was fur-
ther explored by assembling the polymer in
the presence of LUVs. Although CHMP2AAC
CHMP3AC assembly in the presence of SOPC
LUVs had no effect on tube morphology (Fig. 4E),
the presence of DOPS:SOPC LUVs produced
shorter tubes (Fig. 4F), displaying loose helical
coils (Fig. 4G) and cone-shaped tubes that ap-
peared closed at foe narrower aid (Fig. 4H). Thus,
this suggests a mechanism where lipid interac
tion affects CHMP polymerization.

Because modified VPS4 and CHMP3 exert
dominant negative effects on HIV-1 budding
(5, 9, 17, 25) and cytokinesis ( 6 , 18), CHMP2A-
CHMP3 -VPS4 complexes may catalyze a common
step such as membrane fission. The CHMP2A-
CHMP3 polymer presents a membrane binding
topology that is inverse to that of dynamin mem-
brane complexes (26), which catalyze endocytot-
ic vesicle abscission (27). ESCRT-II1 coupled to
VPS4 may exert a similar role in budding pro-
cesses directed away from the cytosol. Thus, we
propose that a helical CHMP2A CHMP3 poly-
mer assembles on the inside of a membrane bud,
which may induce membrane deformation, lead-
ing to constriction and eventually abscission when

coupled to VPS4 activity, the only energy' providing
candidate in the pathway (2, 15) (fig. S10).

References and Notes

L ft. L. William. S. Utbe. Not Rev. Mol. Cell Biol 8. 355
(2007).

2. S. Saksena. J. Sun. T. Chu. S. D. Emr. Trends Biochem.
Sci 32. 561 (2007).

3. J. H. Hurley. Cure. Opin. Cell Biol. 20. 4 (2008).

4. ?. D. »enia«. Virology 344. 55 (2006).

5. J. G Carlton. ). Martin Serrano. Sconce 316. 1908 (2007):
published online 6 June 2007 (10.1126Acience.ll43422>.

6. E. Morita el at.. CMBO ). 26. 4215 (2007).

7. M. Babst. D. J. Katrmann. E. J, Estepa-Sabal T. Meertoo.
S. D. Emr. Dev. Cell 3. 271 (2002).

8. U. K. von Schwedler et at.. Cell 114. 701 (2003).

9. T. Mu/iol et oL Dev. Cell 10. 821 (2006).

10. A. Zamborlini et at., Proc Natl Read. Sa. USA 103.
19140 (2006).

11. S. Lata el at.. ). Mol. Biol 378. 816 (2008).

12. Y. Lin. L A. Kim pier. T. V. Naisrrith. J. M. Uuer.

P. I. Hanson . ). Biol. Chem. 280. 12799 (2005).

13. S. Shim. L. A. Kimpler. P. I. Hanson, Traffic 8. 1068
(2007).

14. P. I. Hanson. R. Roth, Y. Lin. J. E. Heuser, }. Celt Biol
180. 389 (2008).

15. M. Babst. B. Wendland. E. J. Estepa. S. 0. Emr. CMBO J l
17. 2982 (1998).

16. N. Bishop. P. Woodman. Mot. Biol Celt 11. 227 (2000).

17. B. Strack. A. Catistri. E. Popova. H. Gottlinger. Cell 114.
689 (2003).

18. J. D. Dukes. J. D. Richardson. R. Simmons. P. Whitley.
Biochem. }. 411. 233 (2008).

19. L H. Egelman. UUtomkrostopy 85. 225 (2000).

20. Materials and methods are available as supporting
materials on Science Online.

21 M. D. Studied- Brereton et at.. Nature 449. 740 (2007).

22. T. Obita el at.. Nature 449. 735 (2007).

23. A. Scott et at.. CMBO). 24. 3658 (2005).

24. P. Whitley et at..). Biol Chem. 278. 38786 (2003).

25. J. Martin Serrano. A. Yarovoy. D. Perer-Caballero.

P. D. Bieniaa. Proc. Natl Acad. Sci USA 100. 12414
(2003).

26. S. M. Sweftrer. J. E. Hinshaw, Celt 93. 1021 (1998).

27. K. Takei. P. S. McPherson, S. L Schmid, P. De Camilfi.
Nature 374. 186 (1995).

28. We thank X. Siebert (Institut de Biologie Structurale) for
advice on UROX This work was supported by Deutsche
Eorschungsgemeinschaft (SPP 1175) (W.W.). the Agence
Nationale de la Recherche sur le SIDA (W.W.), University
Joseph Fourier (W.W.), the Agence Nationale de la
Recherche (G.S.). the CNRS (GS.). the NIH (grant

Al 29873, H.G). and by postdoctoral fellowships from the
European Molecular Biology Organisation (S.L.) and the
International Human Frontier Science Program
Organisation (S.L.). The EM map has been deposited at
the European Bioinformatics Institute, accession code
EMD 1536.

Supporting Online Material

www.scie ncema g.ot g/cgi/conte nt/f ull/32 1/5894/13547DC1
Materials and Methods
Figs. SI to S10
References

28 May 2008; accepted 24 July 2008
Published online 7 August 2008;

10.112 6/srience.ll61070

Include this information when citing this paper.

A Neoplastic Gene Fusion Mimics
Trans-Splicing of RNAs in Normal
Human Cells

Hui Li, 1 ]inglan Wang, 1 Gil Mor, 2 Jeffrey Sklar 1 *

Chromosomal rearrangements that create gene fusions are common features of human tumors.
The prevailing view is that the resultant chimeric transcripts and proteins are abnormal,
tumor-specific products that provide tumor cells with a growth and/or survival advantage. We show
that normal endometrial stromal cells contain a specific chimeric RNA joining 5' exons of the
JAZFl gene on chromosome 7pl5 to 3' exons of the Polycomb group gene JJAZ1/SUZ12 on
chromosome 17qll and that this RNA is translated into JAZF1-JJAZ1, a protein with anti-apoptotic
activity. The JAZF1-JJAZ1 RNA appears to arise from physiologically regulated trans-splicing
between precursor messenger RNAs for JAZFl and JJAZ1. The chimeric RNA and protein are
identical to those produced from a gene fusion found in human endometrial stromal tumors.
These observations suggest that certain gene fusions may be pro-neoplastic owing to constitutive
expression of chimeric gene products normally generated by trans-splicing of RNAs in
developing tissues.

R ecurrent, specific gene fusions arising
from chromosomal rearrangements are
characteristic features of many neoplasms,
especially those having hematopoietic and mes-
enchymal origins (1-6). In most fusions, recom-

1 Department of Pathology, Yale University School of Medicine,
New Haven, CT 06520, USA. ^Department of Obstetrics and
Gynecology, Yale University Sdiootof Medicine, New Haven, CT
06520, USA.

•To whom correspondence should be addressed. E-mail;
jeffrey.sklar@yale.edu

bination occurs within introns that interrupt the
coding sequences, giving rise to the expression of
chimeric proteins (2). The prevailing view is that
the chimeric proteins resulting from chromosom-
al rearrangements are entirely abnormal and have
neoplastic effects leading to the growth and/or
survival advantage of cells containing them.

An observation that seems at odds with this
view is that chimeric mRNAs identical to those
derived from fusion genes can often be detected
in low abundance by reverse transcription-

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1357

REPORTS

polymerase chain reaction (RT-PCR) of RNA
from healthy tissues (7). The explanation generally
offered for this finding is that specific chromosom-
al rearrangements occur within small numbers of
cells in healthy tissues but that the chimeric pro-
teins generated by than are alone insufficient to
drive substantial clonal expansion.

We have previously described a gene fusion
due to a (7;17Xpl5;q21) chromosomal trans-
location found in about 50% of human endome-
trial stromal sarcomas (ESSs) (8, 9). The fusion
joins the first three exons (from a total of five) in
the gene JAZF1 to the last 15 (from 16) in the
Polycomb group gen tJJAZI/SUZ12. Expression
of the chimeric JAZFl -JJAZI protein in cultured
human embryonic kidney (HEK) 293 cells con-
fers resistance to apoptosis and, when accom
partied by suppression of the unrearranged JJAZI
allele, increased rates of proliferation (9).

We examined normal human endometrial
tissues for possible chimeric JAZFJ -JJAZI
RNA, beginning with endometrial stromal cell
lines. RNA extracted from the immortalized,
normal htunan endometrial stromal cell line
(HESC) (JO) was analyzed for the presence of
JAZFJ -JJAZI chimeric RNA by RT-PCR with
primers containing sense and antisense sequence
flanking the site of joining between JAZFJ and
JJAZJ (JJ). A single amplification product
generated by this reaction was identical in size to
that amplified from human ESSs carrying a
JAZFJ-JJAZJ gene fusion due to the presence
of a t(7;17)(pl5;q21) (Fig. 1A). RT-PCR for
the JAZFJ-JJAZJ RNA in two additional, non-
immortalized primary cell lines derived from
the normal endometrial stroma of two other pa-
tients amplified similarly sized products. Nucleo-
tide sequence analysis of the RT PCR products
from each cell line yielded the same sequence of
nucleotides at the JAZFJ-JJAZJ junction as was
found in RNA of tumors with the gene fusion.
RT-PCR for JAZFJ JJAZI RNA failed to am
plity products from the RNA extracted from a
variety of other epithelial and mesenchymal cell
lines, all of which contained JAZFJ and JJAZI
RNA (fig. SI).

To investigate the specificity of the junction
between JAZFJ and JJAZJ RNA sequences in
the HESC ceil line, we carried out detailed RT
PCR studies on RNA from the HESC cell line
using antisense primers for JJAZJ exon 3 se-
quence paired with six different sense primers for
five exons of JAZFJ. With primers for the first
three exons of JAZFJ, single products were ob-
tained matching the sizes predicted for the join-
ing of JAZFJ exon 3 to JJAZJ exon 2 (Fig. IB).
Similar results were achieved w'hen the sense
primer for JAZFJ exon 3 was paired with six
different antisense primers distributed among the
16 exons of JJAZJ. These results are consistent
with \he JAZFJ -JJAZJ RNA joined at exon 3 and
exon 2 of the respective genes being the only
abundant JAZFJ-JJAZJ RNA in HESC cells.

To determine whether the JAZFJ-JJAZJ
RNA is translated into protein, we performed

Western blot analysis with JJAZI -specific anti
body on protein extracts prepared from HESC
cells. This analysis detected a protein identical in
size to JAZFl -JJAZI protein detected in ESSs
(Fig. 1C and supporting online text).

The detection of JAZFJ-JJAZJ RNA in en-
dometrial stromal cell lines was duplicated by
RT-PCR analysis of RNA extracted from formalin-
fixed, paraffin-embedded tissues from normal
human uteri. JAZFJ-JJAZJ RNA was detected
primarily in endometrium from late secretory and
early proliferative phases of the menstrual cycle
(Fig. 2 A). Ho JAZFJ JJAZJ RNA sequences were
amplified from nonnal myometrium at any phase
of the cycle (fig. S2).

Because of the general association of the
JAZFJ-JJAZJ RNA with endometrium from par
ticular phases of the menstrual cycle, we inves-
tigated the effects of steroid honnones on the
production of the chimeric transcript in HESC

A

cells. Low concentrations of progesterone seemed
to slightly increase amounts of the JAZFJ JJAZI
RNA seen in the absence of added hormone,
whereas both estrogen and, at higher concen-
trations, progesterone suppressed detection of the
chimeric RNA (Fig. 2B). These findings arc
consistent with the results of analyses on endo-
metrial tissue, showing that JAZFJ JJAZI RNA
is present predominantly at the beginning and
end of the menstrual cycle, whai hormone con-
centrations are low.

Because normal endometrium is subjected to
hypoxia and undergoes apoptosis during the late
secretory phase of the menstrual cycle, we in-
vestigated whether hypoxia can induce produc
tion of JAZFJ JJAZJ RNA in HESC cells by
treatment with desferroximine (DFO), which
simulates hypoxic conditions. HESC cells treated
with 250 pM DFO for 8 hours showed increased
amounts of JAZFJ JJAZJ RNA (Fig. 2C). Another,

8 9 1011 12

JJAZI

™

J AZF1 -JJAZI—

JJAZI—

107 kD

81 kD

Fig. 1. Detection of chimeric JAZF1-
JJAZ1 RNA and protein in the HESC
cell line. (A) RT-PCR for the chimeric
junction in th eJAZFl-JJAZl transcript
with primers complementary to anti-
sense and sense sequence in JAZF1
exon 3 and JJAZI exon 2/3 (II). The
figure shows the results of agarose
gel electrophoresis of amplification products. Results of RT-PCR with RNA from an ESS containing
the t(7;17)(pl5;q21) are shown in the lane labeled ESS; results of the RT-PCR procedure with RNA
from the HESC cell line omitting reverse transcriptase are shown in the lane labeled RT; results
without template RNA are shown in the lane labeled dH ? 0. (B) Analyses by RT-PCR for the
specificity of exon joining between JAZF1 and JJAZI RNAs in HESC cells. Lanes 1 to 6 used six
different forward primers at the positions of the downward orange arrows above the diagram of the
JAZFl transcript, paired with a reverse primer indicated by the orange arrow below the JJAZI
transcript Lanes 7 to 12 used the forward primer indicated in green above the JAZFl transcript
paired with six different reverse primers at the positions of the upward green arrows below the
JJAZI transcript (C) Western blot of protein extracts from ESS tissue, the HESC cell line, and HEK
293 cells (as a negative control) for JJAZI and JAZF-JJAZ1 protein with J] AZl-specific antibody.

1358

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

REPORTS

noniinmortalized normal endometrial stromal
cell line, HESC-597, also showed up-regulation
of JAZF1-JJAZ1 RNA when treated with DFO.
Cultures of cells derived from tissue other than
endometrial stroma showed no detectable chi-
meric RNA with DFO treatment. Analysis of the
RNA in DFO-treated HESC cells by a nuclease
protection assay (fig. S3) indicated that some-
what less chimeric RNA was produced in these
cells than in tumor cells containing the JAZF1-
JJAZ1 fiision, consistent with relative amounts of
10 to 35% detected by quantitative RT-PCR. The
nuclease protection assay also confirmed that
detection of chimeric RNA was not a methodo-
logic artifact associated with RT-PCR. Quantita-
tive RT-PCR revealed that treatment of HESC
cells with DFO did not appreciably change the
amount of either JAZF1 ax JJAZl RNA (fig. S4),
suggesting that DFO raises the levels of JAZF1-
JJAZ1 RNA by a mechanism independent of
increased transcription of the two genes.

To investigate whcrhcrJAZFMJAZJ RNA is
produced from a t(7;l 7)(pl 5;q2 1) in HESC cells,
we first showed that this cell line, which had not
intentionally been cloned, had in fact originated
from a single cell immortalized in culture (fig.
S5). We then performed Southern blot analyses
of HESC DNA by using probes that had pre-
viously detected t(7;17)(pl5;q21) rearrangements
in ESSs. No rearranged bands were detected. Cy-
togenetic analysis of numerous metaphase spreads
from HESC cells revealed no abnormalities in
chromosomes 7, 17, or any other chromosome
(Fig. 3A). Similarly, analysis by fluorescence in
situ hybridization (FISH) with pairs of bacterial
artificial chromosome (BAC) probes for DNA
flanking on either side the chromosome 7pl5
breakpoint and separately the I7q2 1 breakpoint
detected no breakage in these regions of the
genome (fig. S5). Additionally, a probe consist
ing of a yeast artificial chromosome (YAC) that
contains DNA spanning the 7pl5 breakpoint

Fig. 2. Detection of JAZFl-JJAZl RNA in endometrial tissues and effects of hormones and hypoxia on
amounts of JAZF1-JJAZ1 RNA in cultured cells. Analyses were performed by RT-PCR, as in Fig. 1A.
Unlabeled panels show results of RT-PCR for p-actin RNA as a control for input RNA. (A) Detection of
JAZF1-JJAZ1 RNA in total RNA extracted from endometrial samples representing various phases of the
menstrual cycle: EP, early proliferative; MP, mid-proliferative; LP, late proliferative; ES, early
secretory; MS, mid-secretory; IS, late secretory. At least two separate uteri were tested for each phase.
(B) Effect of hormone treatment on JAZF1-JJAZ1 RNA in HESC cells. After 2 days of serum starvation,
medium containing serum and no drug (lane 1), 17 p-estradiol at 5 x 10 -8 M (lane 2), progesterone
at 1 x 10“* M, 1 x 10“ 7 M, or 1 x 10" 8 M (lanes 3 to 5), or 17 p-estradiol at 5 x 10" 8 M plus
progesterone at 1 x 10“ 7 M (lane 6) was added to the cells for 24 hours. (C) Effect of DFO treatment
on JAZFl-JJAZl RNA in HESC and HESC-597 cells. No chimeric transcript could be detected in
Ishikawa cells, an endometrial carcinoma line.

showed no splitting of the fluorescent signal (Fig.
3B). Finally, no superimposition of signals was
observed when probes for chromosomes 7pl5
and 17q21 were used together in FISH studies.

To investigde die possibility that (7; 17Xjpl5;q21)
translocations or their equivalents arose in cells in
culture at some point after immortalization, we
subcloncd HESC cells by limiting dilution.
Thirty-seven subclones derived on average from
half a cell per culture were tested for the pro-
duction aiJAZFl-JJAZl RNA. RT-PCR of RNA
from these subclones detected JAZFl-JJAZl
RNA in all clones examined, and for most sub-
clones, the amount of JAZFl-JJAZl RNA in-
creased when DFO was added to the culture (Fig.
3C). Furthermore, analyses of all clones ex-
amined were negative for rearrangements at the
chromosome 7pl5 site by FISH with flanking
probes (Fig. 3D). Ten of these subclones were
also tested for rearrangements at the 17ql 2 site by
FISH, and none of these showed abnormalities.

Given the evidence against DNA recombina-
tion in HESC cells and the precise joining of
sequences at exon boundaries in JAZFl-JJAZl
RNA, we reasoned that the mechanism most
likely responsible for production of this RNA is
trans-splicing of prc-mRNAs for the JAZF1 and
JJAZ1 genes. To test this hypothesis, we prepared
in vitro splicing extracts from the nuclei of HESC
cells. Samples of this extract were mixed with
samples of a nuclear extract from a primary'
rhesus fibroblast cell line RF (12). RT-PCR of
JJAZ1 intron 1 RNA sequence revealed that un-
spliced pre-mRNA was present in the HESC and
RF nuclear extracts (fig. S7). Nucleotide se-
quence analysis of exon 3 in the JAZF1 gene of
RF cells showed two single base-pair sequence
differences from the human JAZF1 gene that
permitted both selective RT-PCR of any RNA
containing cither rhesus or human exon 3 and the
ability to distinguish between the products am-
plified from these RNAs With selective primers
and conditions, amplification of RNA after incu-
bation of mixed extracts yielded products in
which RF JAZF1 exon 3 was joined to exon 2 of
JJAZ1 (Fig. 4B). Sequence analysis of the RT-
PCR products confirmed that the amplified
JAZFl-JJAZl sequences contained exon 3 of
RF JAZF1 (Fig. 4Q. The amount of product
generally increased when tire HESC extract was
prepared from cells cultured with DFO, although
the extent of increase varied considerably among
experiments. No product was obtained from ex-
tracts of HESC cells or RF cells alone, or when
adenosine 5 '-triphosphate (ATP), an obligate
cofactor for splicing, was omitted from the splic-
ing reaction. Similar results were obtained with
extracts from the noniinmortalized endometrial
stromal cell line, HESC-597 (Fig. 4D).

To study the mechanism further and to mle
out the possibility of polymerase switching dur
ing transcription, we carried out the in vitro trans-
splicing assay with HESC nuclear extract mixed
with purified RF RNA. The amount of trans-
spliced product detected was similar to that in the

www.sciencema 9 .or 9 SCIENCE VOL 321 5 SEPTEMBER 2008

1359

REPORTS

1360

assay performed with a mixture of HE SC and RF
nuclear extracts (Fig. 4E). Elimination of all
PCR-detectable traces of DNA from the RF

RNA preparation by treatment with deoxyribo-
nuclease I did not afreet the production of the
trans-spliced product (figs. S7 and S8).

The data presented here are consistent with
trans-splicing of the pre-raRNAs transcribed
from the JAZF1 and JJAZ1 genes in normal

A

It

* *•

At

is i

MM 1!

if

it

At

(**) It

S8 H

o «
-*n

Si

<u

* *

aa }

V

it *4

• A

u

Fig. 3. Absence of the t(7;17)(pl5;q21) in HESC cells. (A) Cytogenetic
analysis of HESC cells. Normal chromosomes 7 and 17 are circled. (B) FISH
analysis of HESC cells with a YAC probe containing DNA spanning the
JAZF1 locus. Arrows point to the intact YAC signal in meta phase
chromosomes. (Lower inset) A representative interphase HESC nucleus;
(upper inset) an ESS control showing splitting of the probe signal. (C) RT-PCR

analysis for JAZF1-JJAZ1 RNA in representative subclones of the HESC cell
line with and without DFO treatment. (D) FISH analysis of three HESC
subclones with two BAC probes, labeled red or green and each containing
DNA sequences that flank the JAZF1 locus on one side or the other. The
juxtaposition of red and green signals indicates no separation of these
sequences. (Inset) Separation of signals in an ESS control. Bars, -10 pm.

A HESC: + + + + -

DFO: - + + + -

ATP: + + - + -
RF: + + + - +

c

AAGATTCAGCCGAAGCT TCGCTGACT
TTCTAAGTCGGCTTCGA AGCCACTCAr^ M fcMl

sequence

AGTCAGCGAG AGCTTC

B

HESC: + + + + -

DFO: - * + + -

ATP: + +•- + -

RF: + + + - +

AAGATTCAGCCGAAGCT TCACTG>
TTCTAAGTCGGCTTCGA AGTGACTGAl

AGTCAGTG ACAGC TTC

Fig. 4. In vitro trans-splicing reactions. (A) RT-PCR analysis for JAZF1-JJAZ1
RNA with a human-specific primer. RT-PCR products from chimeric RNA
were amplified only when ATP was supplied, and the amount of product
increased when the HESC nuclear extracts were prepared from cells pretreated
with DFO. (B) RT-PCR analysis for JAZF1-JJAZ1 RNA with a rhesus-specific
primer. No band was detected in HESC or RF extracts alone, but products
were observed when the two kinds of nuclear extracts were mixed. The
amount of product increased when the HESC extract was prepared from the
cells pretreated with DFO. (C) Sequence analysis of the RT-PCR products

amplified with antisense human- or rhesus-specific primers. Aqua-colored
bases and stars indicate the interspecies sequence differences included in
the species-specific primers; red bases and stars indicate the species-specific
sequence differences detected in the products. (D) RT-PCR analysis for
JAZF1-JJAZ1 RNA in nuclear extracts of the HESC-597 and RF cells with
rhesus-specific primers. Mixed extracts produced detectable signal in the
presence of ATP. (E) RT-PCR analysis for JAZF1-JJAZ1 RNA in HESC nuclear
extracts mixed with purified RF RNA by means of amplification with the
rhesus-specific primer. R, RNA; N, nuclear extract.

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org

REPORTS

endometrial stromal cells and tissues to yield
chimeric products identical to those produced by
a recurrent gene fusion in endometrial stromal tu-
mors. Trans-splicing of noncoding, leader exons
to separately transcribed pre-raRNAs is common
in certain lower eukaryotes, such as protozoa and
nematodes (13-16). However, in vertebrates, only
a few examples of trans-splicing have been de-
scribed (17-25), and most of these involve splic-
ing between pre-mRNAs of the same gene to
generate mRNAs with duplicated exons (1 7-20).
For these reasons, trans-splicing in vertebrates
has sometimes been regarded as a nonfunctional
by-product of a somewhat sloppy splicing sys
tem (26). This conclusion seems inapplicable to
JAZF1-JJAZ1 RNA because the JAZF1-JJAZ1
fusion gene is a recurrent finding in a high frac-
tion of endometrial stromal tissues, and frxsion
genes associated with chromosomal transloca-
tions in cancer have repeatedly been shown to
contribute to the neoplastic phenotype of the
tumors containing them (27, 28). Additionally,
the expression of the JAZF1-JJAZ1 coding se-
quences in cultured cells has demonstrated ef-
fects on cell survival and proliferation (9).
Whether JAZF1-JJAZ1 protein in tissues pro-
vides protection from hypoxia, to which endo-
metrium is subjected during the late secretory
phase and possibly the early proliferative phase
of the menstrual cycle, remains to be determined.

The mechanisms involved in the trans-
splicing of RNAs and the regulation of this pro-
cess are unclear. Juxtaposition of the loci encoding
the RNAs that participate in trans-splicing would
not seem essential because in vitro splicing in
nuclear extracts of RNAs at physiological con-
centrations was found to be efficient These re-
sults also indicate that cotranscriptional splicing
is not an absolute requirement for trans-splicing.
Whether in vivo trans-splicing of RNA tran
scribed from loci that participate in chromosomal
rearrangements predisposes DNA at those sites to
recombination with or without prior intranuclear
colocalization of the loci will require further
investigation.

In view of the regulated trans-splicing be-
tween JAZF1 and JJAZ1 pre-mRNAs in normal
endometrium, the t(7;17Xpl5;q21) found in
ESSs might be considered a mutation that leads
to constitutive production of the JAZF1-JJAZ1
mRNA and its protein product This relation is
similar to that seen in other oncogenic mutations
associated with tumor development, namely, that
mutations lead to overproduction or irreversible
activation of gene products rather than to creation
of “new” genes, as the fusion genes resulting from
many chromosomal translocations and other DNA
rearrangements have generally been thought to be.

If RNA products of fusion genes other than
JAZF1-JJAZ1 also mimic normal products re-
sulting from trans-splicing of pre-mRNAs, it
would explain the ability to frequently amplify
from healthy tissues chimeric RNAs associated
with chromosomal rearrangements in tumors.
Considering the large number of recurrent gene

fusions found in tumors, it would further sug-
gest that trans-spliced RNAs may be relatively
common in normal cells and tissues (support-
ing online text). At a minimum, the finding of the
trans-spliced JAZF1JJAZ1 RNA in normal cells
implies a risk to inferring the presence of chro-
mosomal rearrangements in tissue specimens for
the diagnosis and detection of cancer, especially
in the context of minimal disease. Additionally, it
is possible that drugs designed to target chimeric
proteins produced by neoplastic gene fusions
may have toxicities due to inhibited function of
similar proteins in normal cells.

References and Notes

1. M. A. Pierotti et at. fiat- NatL Acad. Sd. US.A 89.
1616 (1992).

2. F. Mitelman, B. Johansson, F. Mertens. Nat Rev. Cancer
7. 233 (2007).

3. 0. Bernard et at. Oncogene 6. 1477 (1991).

4. P. J. Kourlas el at, froc. Natl. Acod. So'. U.SA 97. 2145
(2000).

5. A. Paidanani et at. Blood 102. 3093 (2003).

6. S. A. Tomlins et at.. Science 310. 644 (2003).

7. S. Jan;. M. Potter. C. S. Babkin. Gene s Chromosomes
Corner 36. 211 (2003).

8. J. I. Koontz et at. froc. NatL Acod. Sci. U.SA. 98. 6348
( 2001 ).

9. H. Li etot. froc. NatL Acod. Sd U.SA 104, 20001 (2007).

10. G. Krikun et at. Endocrinology US. 2291 (2004).

11. Materials and methods are available as supporting
material on Science Online.

12. R. C. Desrosiers et at. ). Virol. 71. 9764 (1997).

13. L Bonen. EASES }. 7. 40 (1993).

14. M. Krause. D. Hirsh, Cell 49. 753 (1987).

15. N. Agabian. Cell 61. 1157 (1990).

16. R. L Sutton, J. C. Boothroyd. Cell 47. S27 (1986).

17. C. Caudevitla et at. froc. Natl. Acad. Sci U.SA. 95.
12185 (1998).

18. A. N. Akopian et at. FEBS Lett. 445, 177 (1999).

19. S. A. Frantz et at. froc. Nott Acod. Sd. U.SA 96. 5400
(1999).

20. T. Takahara. S. I. Kanazu, 5. Yanagisawa. H. Akanuma.

J. Biol. Chem. 275. 38067 (2000).

2L C. Finta. P. G Zaphiropoulos. J. Biot Chem. 277. 5882
( 2002 ).

22. G Flouriot. H. Brand. B. Seraphin, F. Gannon , }. Biot
Chem 277, 26244 (2002).

23. C. Fitzgerald et at. J. Biol Chem. 281. 38172 (2006).

24. 2. Jehan et at. Genome Res. 17. 433 (2007).

25. C. Zhang et at. DNA Cell Biol 22. 303 (2003).

26. T. Maniatis. B. Task. Nature 418. 236 (2002).

27. C. S. Huettner, P. Zhang. R. A. Van Etten, 0. G. Tenen,
Nat Genet 24. 37 (2000).

28. A. T. Look. Science 278. 1059 (1997).

29. Supported by the National Cancer Institute (grant
R01 CA85995) and a generous gift horn the
Bumstein Turnbull family. We thank M. Martel for
assistance in the histologic evaluation of tissue
samples; R. Means for providing the rhesus fibroblast
cell line RF; P. Li for cytogenetic analysis: V. Patriub
for technical assistance; and H. Taylor. J. Steitz.

D. Mi shier. E. Jllu. and A. Krensky for helpful
discussion.

Supporting Online Material

wwn.scie ncemag.org/cgi/dmtent/fuH/32175894/1357/OCl

Materials and Methods

SOM Tent

Figs. SI to S8

References

20 February 2008: accepted 7 July 2008
10.1 12 6/science. 1 156 72 5

Germline Allele-Specific
Expression of TGFBR1 Confers an
Increased Risk of Colorectal Cancer

Laura Valle, 1 Tarsicio Serena-Acedo, 1 Sandya Liyanarachchi, 1 Heather Hampel, 1
llene Comeras, 1 Zhongyuan Li, 1 Qinghua Zeng, 2 Hong-Tao Zhang, 2 Michael ]. Pennison, 2
Maureen Sadim, 2 Boris Pasche, 2 * Stephan M. Tanner, 1 * Albert de la Chapelle 1 *

Much of the genetic predisposition to colorectal cancer (CRO in humans is unexplained.

Studying a Caucasian-dominated population in the United States, we showed that germline
allele-specific expression (ASE) of the gene encoding transforming growth factor-p (TGF-p)
type I receptor, TOFBR1, is a quantitative trait that occurs in 10 to 20% of CRC patients and
1 to 3% of controls. ASE results in reduced expression of the gene, is dominantly inherited,
segregates in families, and occurs in sporadic CRC cases. Although subtle, the reduction in
constitutive TGFBR1 expression alters SMAD-mediated TGF-p signaling. Two major TGFBR1
haplotypes are predominant among ASE cases, which suggests ancestral mutations, but causative
germline changes have not been identified. Conservative estimates suggest that ASE confers a
substantially increased risk of CRC (odds ratio, 8.7; 95% confidence interval, 2.6 to 29.1), but
these estimates require confirmation and will probably show ethnic differences.

T he annual worldwide incidence of colo-
rectal cancer (CRC) exceeds 1 million,
being the second to fourth most common
cancer in industrialized countries (/). Although
diet and lifestyle arc thought to have a strong
impact on CRC risk, genes have a key role in the
predisposition to this cancer. A positive family

history of CRC occurs in 20 to 30% of all pro-
bands. Highly penetrant autosomal dominant
and recessive hereditary forms of CRC account
for at most 5% of all CRC cases (2). Although
additional high- and low penctrance alleles have
been proposed, much of the remaining predis-
position to CRC remains unexplained (J).

www.sciencemag.org SCIENCE VOL 321 5 SEPTEMBER 2008

1361

REPORTS

Abeirations in the transforming growth factor-^
(TGF-p) pathway are heavily involved in CRC
carcinogenesis (4). Although mutations in the
TGF-P type II receptor gene have been explic-
itly associated with CRC (5), the type I receptor
gene ( TGFBR1 ) has received less attention, al-
though there is evidence that a common variant
may be associated with cancer risk (6, 7). We
hypothesized that TGFBR1 is a notable candi-
date for a gene that, when mutated, causes pre-
disposition to CRC or acts as a modifier of other
genes, resulting in a predisposition. Our study
was undertaken to test this assumption.

Given the previously existing evidence that
inherited allele-specific expression of APC acts
as a mechanism of predisposition to familial ad-
enomatous polyposis (8) and of an analogous
mechanism involving DAPK1 in chronic lym-
phocytic leukemia (9), we searched for a similar
association of TGFBR1 with CRC. We hypothc
sized that the putative change might be subtle;
for instance, lowered rather than extinguished
expression of one allele referred to here as ASE,
for allele-specific expression. To test for ASE
in TGFBR1, we chose three single-nucleotide poly-
morphisms (SNPs) (rs3 34348, rs334349, and
rsl 590) in the 3' untranslated region (3TJTR), to
which primer extension with fluorescent nucleo-
tides (SNaPshot) (10) was applied. These three
SNPs are separated by 1916 and 1778 base pairs
(bp), respectively, yet they exhibit total linkage
disequilibrium.

Among a total of 242 patients with micro-
satellite instability (MSI)-ncgative CRC (10), 96
(39.7%) were heterozygous for the three 3TJTR
SNPs, of whom 12 showed ASE variation ratios
higher than 1.5, whereas no patient showed ratios
below 0.67. Forty-nine additional cases were
heterozygous for one further SNP (rs7871490)
located in the 3UTR that was not in strong
linkage disequilibrium with the above three
markers, and 17 out of 49 (17/49) had ASE val-
ues higher than 1.5. Thus, 29 out of 138 (21%)
informative CRC patients showed ASE in the
TGFBR1 gene. Three additional cases had bor
derline values (fig. SI and table SI).

DNA samples from the blood of healthy
Columbus, Ohio-area controls (195 individuals)
(10) were genotyped for the four SNPs. One
hundred and nine (55.9%) were heterozygous,
and ASE analysis in 105 of them revealed ratios
ranging between 0.72 and 3.25 (fig. SI). Only
three controls showed ratios above 1.5. Our
results in both the CRC patients and controls
suggest that the degree of ASE is a quantita-
tive trait (Fig. 1). Differences in the degree of

'Human Cancer Genetics Program, Comprehensive Cancer
Center, the Ohio State Univeisity, Columbus, OH 43210, USA.
dancer Genetics Program, Division of Hematotogy/Oncology,
Department of Medicine and Robert H. Lurie Comprehensive
Cancer Center, Feinberg Schodl of Medicine. Northwestern
University, Chicago, U 60611, USA
•To whom correspondence should be addressed. E-mail:
bpasche@nonhwestem.edu (B.P.1; Stepban.Tanner@osumc.
edu (S.M.T.); Albert.delaChapelle@osumc.edu (Ad.l.C.)

ASE between patients and controls showed a
P value of 0.1208 when a Wilcoxon rank sum
test was applied and a P value of 0.0207 when a
permutation test (100,000 permutations) was
applied.

At this stage, it is not possible to determine
whether the degree of predisposition to CRC is
proportional to the degree of ASE or whether
there is a threshold value that separates “abnor-
mal” values that predispose to CRC from “nor-
mal” values that do not A ratio of 1 means that
both alleles are equally expressed, whereas a ratio
of 1.5 means a 33% difference, as docs a ratio
of 0.67. To define a cutoff point, we applied
receiver operating characteristic (ROC) analysis,
which estimates the sensitivity and specificity
of cutoff points. As shown in table S2, the
value of 1.5 maximizes both characteristics,
providing the highest Youden’s index. When a
cutoff of 1 .5 was used, the P value comparing
cases and controls was 7.655 x l(f 5 . Although
there is no overall need to define a firm cutoff
point, we used the value of 1.5 to categorize CRC
cases and controls into ASE and non-ASE. In
order to determine whether the observed ratios
falling outside this range represent an increase or
decrease in the transcript of one allele, a reverse
transcription polymerase chain reaction (RT PCR)
experiment was performed, taking advantage
of hybrid clones monoallelic for chromosome 9
created from two individuals with ASE (patients
1 and 26, table SI). Each of the four hybrid
clones contained either the maternal or paternal
copy of chromosome 9, plus the mouse genome
(10). As shown in Fig. 2 A, ASE determination
in the diploid samples indicated that the expres-
sion of one allele (a) was reduced as compared
to that of the other allele (b). In the four mono-
allelic hybrid clones, the dcnsitometric values of
the RT PCR of human TGFBR1 were compared
with the corresponding values for mouse Gpi
(10). One allele (a) showed reduced expression
in both patients. These experiments support the
notion of lowered expression of one allele, and
in both patients the same allele was affected
(Fig. 2, A and B).

Fig. 1. TGFBR1 ASE distribution in
138 CRC patients and 105 controls
studied by SNaPshot The ASE cutoff
value of 1.5 chosen to categorize the
cases is indicated, together with its as-
sociated P value obtained from compar-
ing the proportions of cases (29/138)
and controls (3/105) above the indi-
cated value.

To assess the effect of ASE on TGF-P signal-
ing, lymphoblastoid cell lines from four ASE
patients and four non-ASE healthy controls were
exposed to TGF-P (10), which binds TGFBR2 and
leads to the formation of the TGFBR2/TGFBR1/
TGF-P hctcromeric complex. We observed dif-
ferences in levels of phosphorylated SMAD2
(pSMAD2), an important downstream effector
and surrogate marker of TGF-p signaling (11, 12).
There were constitutive differences in pSMAD2
expression between ASE patients and non-ASE
controls in the absence of exogenously added
TGF-P (time 0; Fig. 3A). Differences in pSMAD2
levels became more pronounced upon exposure
to TGF-p. These differences were observed at
low TGF-p concentrations (<5 pM) (Fig. 3B)
and occurred in four out of four ASE cases as
compared to non-ASE controls.

It has been shown that phosphorylation of
SMAD3 is an essential step in signal transduc-
tion by TGF-p for the inhibition of cell prolif-
eration (13). Furthermore, SVwarfi-deficicnt mice
are prone to developing colon cancer (14, 15).
To assess the impact of TGFBR1 ASE on ihc
phosphorylation of SMAD3, we used an antibody
targeting the Ser 42W25 site in SMAD3 (10, 16).
Constitutive levels of pSMAD3 were detectable
in the lymphoblastoid cell lines of three ncm-ASE
controls, whereas pSMAD3 was barely detect-
able in one ASE case (Fig. 3C). Exposure to
TGF-P did not result in any detectable increase
in pSMAD3 in the lymphoblastoid cell lines of
the ASE patients. The pSMAD2 and pSMAD3
results indicate that patients with ASE exhibit
decreased SMAD-mediated signaling when com-
pared with non-ASE controls.

A GCG trinucleotide variable number of tan-
dem repeat polymorphism occurs in exon 1 of
TGFBR1. The most common allele contains nine
repeats leading to a stretch of nine alanines (9A)
in the signal peptide of the receptor protein. The
second most common allele has six repeats (6A)
and occurs in approximately 14% of all indi-
viduals in most Caucasian populations (6). The
6A allele has been associated with a low-level
but statistically significant predisposition to sev-

0.75

Controls Cases

n=105 n=138

1362

5 SEPTEMBER 2008 VOL 321 SCIENCE wvwv.sciencemag.org

REPORTS

CRC patient 1

'£k

gDNA cDNA

CRC patient 26

Monochromosomal • Monochromosomal
hybrids 1 : hybrids 26

Allele 8 Allele b ; Allele a Allele b

gDNA

Fig. 2. ASE determination
in two ASE CRC probands.
(A) ASE detection in blood
DNA by SNaPshoL The ASE
ratio was calculated by nor-
malizing the ratio between
the peak areas of the two
alleles in cDNA with the same
parameters in genomic DNA
(gDNA). In both examples,
the transcript from the a al-
lele is reduced with respect
to that from the b allele. (B) Semiquantitative RT-PCR of the cDNA from monochromosomal
hybrids of the same two patients. Human TGFBR1 expression (amplicon size 135 bp) was as-
sessed and mouse Gpi was used as a control (176 bp). The values shown below the gel represent
the ratios of the densitometric values of human TGFBR1 versus mouse Gpi, showing reduced
expression of human TGFBR1 in the hybrids that contain the a allele.

Mouse Gpi
Human TGFBR1

TGFBR1 / Gpi
Allele b / allele a

0.60 0.8
1.48

| 0.53 0.8

1.51

C-1 (ASE ' 03)
pSMAD2 jHHHI
SMAD2 rT»| -

(3-actin

C-1 (ASE 1.03)1 P-1 (ASE 3.95)

Time o

0 111 4h Bh 16h

pSMAD2

P -actin
Time

C-3 (ASE 1 04)

P-14 (ASE 1.74)

1 0 IK «h 8h l«jh

0 in -mi » i6n

TGF-P

TGF-0

Histone I^HHI
Time 0 1h 16 h

[ 0 in ten

TGF-P

TGF-p

C-2 (ASE 0.99)

P-5 (ASE 2.24)

Time 0 ih ion 1

I 0 Ih 16h

TGF-p

TGF-P

C-3 (ASE 1 04)

P-14 (ASE 1.74)

pSMAD3^^^^||

Time 0 ih ten I

I 0 ik ien

TGF-p

TGF-P

pSMAD2
SMAD2
P -actin

C-4 (ASE 1.13)

P -26 (ASE 1.54)

i m

_ _ n rw

) 0 5 25 50 100

0 5 25 50 100

Fig. 3. Analysis of SMAD-mediated TGF-p sig-
naling in lymphoblastoid cell lines from ASE
CRC patients and non-ASE healthy controls. (A)
SMAD2 and phosphorytated S /VLAD 2 (pSMAD2)
expression were assessed by Western blotting
in lymphoblastoid cell lines from ASE patients
(P-1, P-5, and P-14) and non-ASE controls (C-1,
C-2, and C-3), after exposure to TGF-p (100 pM)
at various time points from 0 to 16 hours (h)
and using p-actin as a loading control. In all
three ASE cases, less constitutive pSMAD2 was
observed than in non-ASE controls. The differences in pSMAD2 expression between ASE and non-ASE
cell lines were further enhanced after exposure to TGF-p. (B) SMAD2 and p-SMAD2 expression 1 hour
after exposure to different TGF-p concentrations. The effect shown in (A) also occurs at low concentra-
tions of TGF-p (5 pM). (O pSMAD3 detection in nuclear extracts from three ASE patients and three non-
ASE controls after exposure to TGF-pi. The three non-ASE lymphoblastoid cell lines had pSMAD3
expression in the nucleus, whereas nuclear pSMAD3 expression was undetectable in two ASE cases (P-1
and P-14) and barely detectable in one case (P-5).

TGF-0 1 hour

eral forms of cancer (/ 7-20). Recent studies sug- ^
gest that the assoc iaaion of 6A with colon cancer
is either weak [odds ratio (OR) 1.2, 95% confi-
dence interval (Cl) 1.01 to 1.43] ( 17) or border-
line significant (OR 1.13, Cl 0.98 to 1.30) (21).

We typed this polymorphism in all 242 CRC
cases studied by us and found 9 A/9 A in 197,
9A/6A in 40, 6A/6A in 4, and 1 tailed (table S3).
There were clearly more 9A/6A heterozygotes
among the patients with ASE (14/29) than in
those without ASE (22/108) ( P = 0.0052, chi-
square test). We tentatively concluded fiom these
data that die 6A allele is probably in linkage
disequilibrium with one of the putative muta-
tions that causes ASE, but 6A is not in itself
causative of ASE.

All 29 patients showing ASE and three pa-
tients with borderline ASE values (1.49, 1.49,
and 1 .46) (n = 32 patients) were studied for ge-
netic changes occurring in the germ line. By
sequencing of all nine exons, 2 kb upstream of
exon 1, and the entire 3TJTR (10), a single se-
quence change in the coding exons was iden-
tified in patient 30, consisting of a coding DNA
1204 T— *A (c.l204T>A) missense change in
exon 7 that changes a tyrosine to asparagine _
(p.Tyr401Asn). Its pathogenicity is currently
being assessed. Several changes, all previously
reported as polymorphic, were identified in the
3'UTR and promoter regions. In three patients, a
deletion (del) of two bases (c. l -1782_1783delCA)
at 1783 bp upstream of exon I was identified in
a repetitive sequence of short interspersed nu-
clear elements. Multiplex ligation-dependent
probe amplification (10) did not suggest any
large rearrangements, deletions, or duplications
of exons. In a study of promoter methylation,
none of the comparisons of germline methyla-
tion status between ASE and non-ASE cases
and ASE cases versus controls were significant
(supporting online material text and table S4).
Thus, germline promoter methylation is unlikely
to play a role in ASE.

We hypothesized that changes occurring in
noncoding regions of the gene could be respon-
sible for the reduction in expression. To fully

study this possibility, overlapping fragments
of 1.7 to 10 kb were amplified by long-range
PCR, cloned, and sequenced. In all, approximate-
ly 96.5 kb covering the whole gene and 3'UTR

(49 kb), 35 kb upstream of exon 1 (up to the
next gene COL15A1), and 12.5 kb downstream
of the 3TJTR (Fig. 4) were fully sequenced in
the four monochromosomal hybrids (patients

www.sciencema 9 . 0 rg SCIENCE VOL 321 5 SEPTEMBER 2008

1363

REPORTS

1 and 26) and in diploid DNA from four other
ASE patents (patients 5, 11, 14, and 21) (10).
Our sequencing strategy allowed us to determine
the phase of every change within each amplicon
and over larger regions when at least one change
occurred in the overlapping fragments. In all,
25 and 104 changes were identified in the down-
regulated alleles of patients 1 and 26, respec-
tively, whereas 3 1 and 6 changes were detected
in their wild-type counterparts. Diploid DNA
from the four patients harbored 61, 37, 33, and
135 changes, respectively.

Excluding changes known to be present in
the wild-type alleles, 140 changes were iden-
tified in the down-regulated alleles. Only the
c.l-1782_1783delCA change stood out as a
candidate mutation. It occurred in 3/29 (103%)
ASE patients, in 0/3 ASE controls, in 1/51 (2%)
non-ASE CRC patients, and 1/81 (1.2%) non
ASE controls. In summary, these investigations
did not uncover the genetic changes causing
ASE.

Genotyping of most changes identified by
sequencing was carried out in all available ASE
CRC patients, including borderline cases (n -
31), and in 55 non-ASE CRC patients. Construc-
tion of haplotypcs from the available genotype
and haplotype data was performed with PHASE
v.2.1.1 (10). In all, 60 polymorphisms cover-
ing 73.5 kb (from 12 kb upstream of exon 1 to

12.5 kb downstream of the 3'UTR) were used
for haplotype inference (table S5). For all ASE
and non-ASE patients, the program was run
with 1000 permutations with overlapping 10-SNP
sliding windows. Haplotype frequency distri-
butions in ASE and non-ASE populations showed
significant differences in a genomic region
covering the area between the 3' end of intron
3 to ~5 kb downstream of the 3' end of the
UIR (Fig. 4).

The group of patients carrying the minor al-
lele for the three 3'UTR SNPs in linkage dis-
equilibrium (group 1) was very different from
the other group derived from the study of SNP
rs7871490 (group 2). Haplotype analysis was
performed separately in the two groups, using
50 and 21 SNPs, respectively. In group 1 (n =
53), cme major haplotype for the affected alleles
was present in 1 1/14 (78.6%) of ASE but also in
22/39 (56.4%) non-ASE patients (Fig. 4). For
group 2 (n - 33), another major haplotype for
the affected allele was present in 14/17 (82.4%)
of ASE and in 1/16 (6.3%) of non-ASE patients
(Fig. 4). Fisher’s exact test to compare haplo-
type proportions showed P values of 0.2031 and
1.260 x 10 -5 for groups 1 and 2, respectively.
The 6A allele of the 9A/6A polymorphism
occurred in the ASE haplotype in all 14 cases of
group 2, but not in group 1, where all ASE cases
except one were homozygous for the 9A allele.

In search of somatic changes in line with
Knudson’s two-hit hypothesis, loss of heterozy-
gosity (LOH) analyses as well as a search for
somatic mutations in the coding sequences of
the gene were performed in DNA from the
tumors of 26 ASE patients. Using the described
threshold (10), 6 cases out of 26 showed LOH.
In three of these six cases, the wild-type allele,
the one with normal expression in blood, was
lost or reduced, whereas in the other three cases,
the allele showing gcrmlinc ASE was lost
Exon-by-exon sequencing of the entire gene
in tumors from 26 ASE patients revealed so-
matic changes in three tumors that were not
found in blood DNA. The mutations were:
c.634G>A (p.Gly212Asp) in one tumor and
c.682 685delAAG (p.Ghi228del) in two tumors.
These mutations occurred in exon 4, which
encodes the kinase domain of the protein.
LOH analyses and exon 4 sequencing in 49 tu-
mors of CRC patients without ASE showed that
none of these tumors had evidence of somat-
ically aoquired mutations, and five showed LOH
(table S3). Fisher’s exact test comparing propor-
tions of LOH and mutations between ASE and
non-ASE cases showed P values of 0.1708 and
0.0355, respectively. The occurrence of somatic
mutations in ASE cases but not in controls
supports the role of TGFBR1 as a tumor sup-
pressor gene. On the other hand, the fact that

. Soquencod in 6 patients 96 5 Kb

c1-1782_1783d0tCA 9A/6A

rs334348 ] \

rs787 M90\ '•

1*334349 \

is 1590

Scale
S Kb

SNPs used for haplotype inference

li e i n « i i n h i

H H-

-H— I 1 ti l I \\ I

Significance of haplotype
frequency dstributions in 31 ASE
and 55 non-ASE patients

c

Inferred risk haplotypes by groups

Group

p=0 660 t

p=0 717

DO 536

p=0 701 , p=0 076

, D-0 563 ,

„ P=0.421 „

„ asuaa *

■p-g,9?s „

Haplotype (48 SNPs) of affected allele shared by 11/14 ASE Bnd 22/39 non-ASE CRC patients
GCT AT GAAGAGCATA GTTC A C ACAGT G GCAA GTGGAG T ATCGCTA CTA

Group 2

Haplolype (20 polymorphisms including 9A/6A) of affected alete shared by 14/17 ASE and Vi 6 non-ASE CRC patients

T3T

Fig. 4. (A) Diagram of the TGFBR1 genomic region. The uppermost line depicts the 96.5-kb region sequenced in six ASE
patients (four monochromosomal hybrids and four dipbid DNAs). Shown are the locations of the 2-bp CA deletion

upstream of exon 1, the 9A/6A polymorphism in exon 1, and the four SNPs in the 3'UTR used for ASE determinations. (B) Q ba mimo ot mo 9A/6A polymorph*™
Locations of the 60 SNPs used for haplotype inference in ASE (n = 31) and non-ASE (n = 55) CRC patients. The arrowed

shorter lines each depict a 10-SNP overlapping window. P values indicate the significance of differences in haplotype distribution between ASE and non-ASE
individuals. (C) Two major haplotypes identified in ASE patients are shown.

1364

5 SEPTEMBER 2008 VOL 321 SCIENCE wvwv.sciencemag.org

REPORTS

LOH affected the ASE allele as often as the
wild-type allele could indicate random losses.

The cohort of MSI-negative CRC patients
had been deliberately enriched in familial cases
(JO). In the cohort of 138 patients with available
ASE values, 59 out of 136 (43.4%) were fa-
milial according to the criteria indicated above,
and family information was not available in two
cases. Among the cases showing ASE, 53.6%
were familial (table S3). The proportion of ASE
was higher among familial than non familial
cases: 15/59 (25.4%) familial cases versus 13/77
nonfamilial cases (16.9%). A chi-square test to
compare proportions showed that this difference
was not statistically significant (P = 0.3 14).

The above data suggest that ASE contrib-
utes somewhat more to familial than to spo-
radic CRC but do not allow its inheritance to
be assessed. If ASE is regularly inherited as a
dominant trait, the expectation is that 50% of
first-degree relatives (FDRs) also have ASE.
Data firom four families that are informative in
this regard are shown in fig. S2. In all, among
1 1 FDRs, ASE was greater than 1 .5 in 4, bor
derline in 2 (ASE values 1 .40 and 1 .44), and
low in 5. There was no instance of ASE being
incompatible with Mendelian dominant inher
itance. In all four families, co-segregation of
ASE with the inferred risk haplotypc, represent
ing the down-regulated allele, occurred. The
highest Kong and Cox nonparamctric LOD
score was 1 .25, with a P value of 0.008 (non-
parametric z score = 4.12; P value = 0.00002).
Among the four to six ASE-positive FDRs, two
had CRC, one had endometrial cancer and a
tubular colonic adenoma, one had prostate can-
cer, and another had multiple polyps in the
colon and rectum (table S6). Although fragmen-
tary, these data suggest dominant inheritance of
ASE with incomplete penetrance of CRC in
ASE carriers.

There is indirect evidence to support the
notion that ASE of TGFBRJ contributes to CRC
development. The TGF-p pathway is strongly
involved in the carcinogenesis of colon and
other cancers, and its signaling is dependent on
the integrity of both of its receptors (TGFBR1
and TGFBR2) (22, 23). In a comprehensive
study of CRC tumors, somatic mutations oc-
curred with high frequency in 69/13,023 genes.
Among these 69 genes were TGFBR2, SMAD4,
SMAD2, and SMADS , attesting to the impor

tancc of the TGF-P pathway in CRC (24). There
is rapidly increasing evidence that subtle var-
iations in gene expression play central roles not
only in development in various organisms but
also in human disease (8, 9, 25). Linkage anal-
ysis of a cohort of sibling pairs concordant or
discordant for colorectal carcinoma or adeno-
ma highlighted a region in chromosome 9q22
31 (26). Subsequently, borderline significant
linkage to the same region was observed in fam-
ilies segregating colorectal cancer or adenoma
without microsatellite instability (27, 28). This
evidence is compatible with, but in no way
proves, a role for TGFBRJ.

We were unable to determine what mecha-
nism causes ASE. The haplotype data support
the implication of ancestral mutations for most
ASE patients. Moreover, the elusive genomic
change causing ASE is likely to occur in cis, but
the data do not exclude the possibility that ASE
arises as a result of trans acting genes that pref-
erentially affect the risk haplotypes. Such genes
could well be RNA genes as predicted earlier
(29). Very recently, the existence of extensive
quantitative trait loci for gene expression was
documented in two large studies (30, 31).

How common is ASE of TGFBRJ ? Using
our definition, it occurred in 29/138 tested CRC
patients (21%) and in 3/105 tested controls (3%).
In the extreme, if none of the non-informalive
CRC cases had ASE, the frequency would be
29/242 (12%), and for the controls, 3/195 (1.5%).
Because not all individuals are informative
(heterozygous for a transcribed SNP), the true
frequency in cases and controls cannot be pre-
cisely assessed at present. Using the above alter-
native numbers, we can calculate the OR of
CRC in carriers of ASE. In the first scenario, the
OR is 9.0 (Cl 2.7 to 30.6), and in the conserv-
ative one, OR is 8.7 (Cl 2.6 to 29.1).

What proportion of all CRC is attributable
to ASE of TGFBRJ ? From the available data
of the present case-control study, we estimated
the population attributable risk (PAR). If ASE
occurs in 21% of cases and 3% of controls,
the estimated PAR is 18.7% (Cl 10.8 to 25.8).
If ASE occurs in 12% of cases and 1.5% of
controls, the estimated PAR is 10.6% (Cl 6.0
to 14.9). These numbers arc estimates, represent
ing the Caucasian-dominated population of
central Ohio, and are heavily dependent on the
relevant allele frequencies, which may show

strong inter ethnic variation. We nevertheless
conclude that ASE of TGFBRJ is a major con-
tributor to the genetic predisposition to CRC.

References and Notes

1 0. M. Parkin, F. Bray. J. Fertay. P. Pisani, CA Cancer
J. Clin. 55. 74 (2005).

2. A. de la Chapelle. Nat. Rev. Cancer 4. 769 (2004).

3. N. M. lindor et oi. JAMA 293. 1979 (2005).

4. P. M. Siegel J. Massague. Nat Rev. Cancer 3. 807
(2003).

5. S. Markowitz et al„ Science 268. 1336 (1995).

6. 8. Pasche et al.. Cancer Res. 58. 2727 (1998).

7. Y. Xu. B. Pasche. Hum. Mol. Genet 16. R14 (2007).

8. H. Yan et at.. Not Genet. 30. 25 (2002).

9. A. Raval et at. Celt 129. 879 (2007).

10. Materials and methods are available as supporting
material on Science Online.

11. J. Massague. R. R. Gomis. FEBS Lett 580. 2811 (2006).

12. j. Massague. Mol. Ceil 29. 149 (2008).

13. X. Liu et at.. Rroc. Nati Acad. So. US. A 94, 10669
(1997).

14. Y. Zhu. J. A Richardson. L F. Parada. J. M. Graff. Celt 94.
703 (1998).

15. N. M. Sodir et at.. Cancer Rei. 66. 8430 (2006).

16. G Sekimoto et at.. Cancer Re s. 67. 5090 (2007).

17. B. Pasche et at.. J. Clin. OncoL 22. 756 (2004).

18. 8. Pasche ef oi. Cancer Res. 59. 5678 (1999).

19. B. Pasche et oi. JAMA 294. 1634 (2005).

20. Y. Bian et at.. J Cm. Oncol. 23. 3074 (2005).

21. J. Skoglund et oi, Ctin. Cancer Res. 13. 3748 (2007).

22. B. Bierie. H. L. Moses. Cytokine Growth Factor Rev. 17.
29 (2006).

23. J. Massague. S. W. Blain. R. S. lo. Cell 103, 295 (2000).

24. T. Sjoblom et at.. Science 314. 268 (2006).

25. H. Yan. W. Zhou. Curt. Opin. Oncol. 16. 39 (2004).

26. G L Wiesner et oi. Rroc. Natl. Acad Sd. USA 100.
12961 (2003).

27. Z. E. Kemp et at. Cancer Res 66. 5003 (2006).

28. J. Skoglund et at.. J. Med Genet 43. e7 (2006).

29. M. Morley et at.. Nature 430. 743 (2004).

30. H. H. Goring et at.. Not. Genet 39. 1208 (2007).

31. B. E. Stranger et oi. Nat Genet. 39. 1217 (2007).

32. We thank H. He. K. Sotamaa, and W. Frankel
for help. This work was supported by NIH grants
CA67941. CA16058. CA112520. and CA108741;
grants from the Walter S. Mander Foundation. Chicago.
II; and the Jeannft M, Littlefield- American Association
for Cancer Research Grant in Metastatic Colon Cancer
Research. LV. was supported by a fellowship from the
Fundacion Ramon Areces.

Supporting Online Material

www.scie ncema g.org/cgt/conte nt/futl/11 59397/00

Materials and Methods

SOM Text

Figs. Si and S2

Tables SI to S6

References

21 April 2008; accepted 5 August 2008
Published online 14 August 2008;

10.1 126/science. 1159 39 7

Include this information when citing this paper.

www.sciencemag.org SCIENCE VOL 321

5 SEPTEMBER 2008

1365

New Products Focus: Mass Spectrometry

AAAS/Sc/ence Business Office

LC-MS/MS System

The TSQ Vantage is a new liquid chromatography-mass spectrometry/mass spectrometry
(LC-MS/MS) system that offers increased sensitivity without increased noise for better reproduc-
ibility, accuracy, and precision in quantitative analysis. Technical breakthroughs have improved the
TSQ Vantage’s signal-to-noise performance, providing better reproducibility and precision in the
quantitative analysis of small molecules, biomolecules, and peptides. The instrument offers up to
10 times the signal-to-noise ratio compared with the TSQ Quantum series. The TSQ Vantage’s new
S-Lens ion optics system features a novel electrostatic field technology to capture virtually every
ion and efficiently transfer it into the HyperQuad quadrupole mass analyzer. The S-Lens design is
an advance over high-pressure, skimmer-based ion source designs because it eliminates mass
discrimination and lowers the gas load on the expensive turbo-molecular pumps.

Thermo Fisher Scientific
For information 800-532-4752
www.thermo.com/tsqvantage

Glycopeptide Analysis Software

The new version of the innovative mass spectrometry (MS) data
analysis tool, SimGlycan, can analyze glycopeptides in addition to
released glycans. SimGlycan matches MS/MS spectra with its own
comprehensive, robust, and annotated database to predict the
structure of glycans. It generates a list of all the probable glycans that
are close to the experimental data, saving time and laborious work.
Each structure is scored to help the user judge which results closely
match the experimental data. Alongside the probable glycan structure,
the software makes available information such as the glycan class,
reaction, pathway, and enzyme. Users can resolve glycan structures
in glycopeptide molecules by specifying the sequence or mass of the
attached peptide moiety.

Premier Biosoft International
For information 650-856-2703
www.PremierBiosoft.com

Surfactant

ProteasMAX Surfactant is designed to enhance the enzymatic
performance of trypsin and other proteases in preparation for analysis
by mass spectrometry or liquid chromatography. The surfactant
provides a more complete digestion prior to protein analysis, resulting
in more accurate data in a shorter time and reduced risk of sample
degradation. The new surfactant exposes cleavage sites usually
inaccessible because of secondary or tertiary structure and can
increase sample recovery from gels. It can reduce standard protein
digestion time from overnight to just three hours. It also enhances
protein solubilization at room temperature, which means that high
temperatures and precipitation can be avoided, even for complex
proteins. Because ProteasMAX degrades during the digestion
reaction, researchers can proceed directly to analysis of peptides by
mass spectrometry without additional deactivation steps.

Promega

For information 608-274-4330
www.promega.com

Cell Lysis Kit

The ProteaPrep Cell Lysis Kit, Mass Spec Grade, is designed for the
efficient recovery of purified protein lysates from biological samples. The
kit is a proprietary mixture of salts, glycerol, and an acid-labile surfactant
optimized for efficient solubilization, extraction, and recovery of
proteins during cell lysis. Because the kit contains no lysozyme or harsh

detergents, it permits recovery of cell lysates free from artifactual protein
contaminants and promotes efficient sample protein solubilization. It is
optimal for mass spectrometry, enzymatic digestions, two-dimensional
gel electrophoresis, and protein quantitation procedures.

Prolea Biosciences

For information 877-776-8321

www.proteabio.com

Tandem Quadrupole MS

The Waters Xevo TQ MS System is an advanced tandem quadrupole
mass spectrometer that allows scientists with varying levels of mass
spectrometry (MS) expertise to quickly produce high quality data.
The system features IntelliStart, a new technology that simplifies
instrument setup and troubleshooting. It automatically ensures the
system is ready for use by performing mass calibration, setting MS
resolution, generating compound-specific MS methods, and optimizing
source conditions. The instrument is equipped with a unique collision
cell that can be operated in conventional T-Wave-1 enabled mode or
in a new ScanWave-enabled mode. ScanWave improves the duty cycle
and enhances the full-scan capability to meet the demands of complex
analyses, allowing users to more easily confirm the identities and
structures of targeted analytes.

Waters

For information 508-482-2614
www.waters.com

High-Resolution Time-of-Flight MS

The maXis electrospray ultrahigh-resolution time-of-flight mass
spectrometer offers high resolution of 40,000-60,000 FWHM over
a broad mass range and mass accuracy typically between 600 and
800 parts per billion, at speeds of up to 20 full spectra per second.
The maXis offers 20 Hz full spectra acquisition at high mass
resolution for high-speed liquid chromatography, a high dynamic range
of five orders of magnitude for trace detection in complex mixtures,
and sub-parts-per-million mass accuracy in both mass spectrometry
(MS) and MS/MS mode. The instrument makes high-resolution
mass spectrometry over a broad mass range compatible with ultra
performance liquid chromatography.

BrukerDaltonics

For information 978-667-9580

www.bdal.com

Electronically submit your new product description or product literature information! Go to www.sciencemag.org/products/newproducts.dtl for more information.
Newly offered instrumentation, apparatus, and laboratory materials of interest to researchers in all disciplines in academic, industrial, and governmental organizations are
featured in this space. Emphasis is given to purpose, chief characteristics, and availabilty of products and materials. Endorsement by Science or A AAS of any products or
materials mentioned is not implied. Additional information may be obtained from the manufacturer or supplier.

1374

5 SEPTEMBER 2008 VOL 321 SCIENCE www.sciencemag.org/products